正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

I coded a Speech-to-Text model from scratch. 𝐇𝐞𝐫𝐞 𝐢𝐬 𝐭𝐡𝐞 𝐛𝐥𝐨𝐠 𝐟𝐨𝐫 𝐭𝐡𝐞 𝐬𝐚𝐦𝐞: No APIs. No pre-trained models. Just PyTorch, an A100 GPU, and hours of debugging. This started months ago. I wanted to understand how machines hear. Not surface-level understanding. I wanted to build the whole thing... myself. So I built it piece by piece: autoencoders, VAEs, VQ-VAEs, Residual Vector Quantization, and CTC loss. Each one took days to get right. Trained for 3 hours on 13,100 audio clips. Got complete garbage. Changed the tokenizer from BPE to character-level. Rechecked everything. Asked AVB who built STT models before. His answer: these models are tricky to train and need days of compute, not hours. Cut the dataset to 200 clips. After 2 hours, actual words appeared. Overfitted? Absolutely. But watching noise turn into recognizable English was satisfying. I have made a blog about this as well so you can learn about the same and my process - Audio fundamentals and waveform representation - Why attention breaks on raw audio - Convolutional downsampling - Transformer encoder with positional encoding - Vector Quantization, straight-through estimator, and RVQ - CTC loss and greedy decoding - Full training loop with VQ loss warmup - What went wrong and what finally worked Resources: - Blog: - Code: More Resoures CTC loss AVB videos SoundStream Paper LJ speech dataset wav2vec paper RVQ blog Next up: I've already trained two TTS architectures from scratch. Video post about those coming soon. But first, I'm dropping a visual breakdown of Vision Transformers, covering how they work and how to fine-tune them. Follow me Mayank Pratap Singh you're into audio deep learning. Repost so others can find thisshow more

Mayank Pratap Singh

1,814 subscribers

51,382 次观看 • 3 个月前 •via X (Twitter)

教育科学技术

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

What if your voice AI could interrupt you the moment it figured out your question - sometimes even before you finished asking it? Last week, I sat down with Neil, CEO of Gradium and co-founder of Kyutai , to talk about the future of speech-to-speech models and why he believes today's cascaded voice systems will soon look "archaic and brittle." Some highlights from our conversation: 🎯 How Kyutai built Moshi—a full duplex conversational AI with "negative latency"—in 6 months with just 4-6 people (while big tech teams had 10-20x the resources) 🧠 Why speech-to-speech models lose intelligence compared to their text counterparts (and what's being done about it) 📱 Pocket TTS: The first voice cloning model that runs on your phone's CPU—not GPU, CPU 🤖 Why robotics and spatial audio represent the next frontier (hint: current voice systems completely break in these environments) 👶 The efficiency gap: Babies learn to speak fluently from <5,000 hours of audio. Current models train on millions of hours. We're doing something wrong. My favorite vision from Neil? The first truly contrarian AI that interrupts you mid-sentence to tell you why you're wrong. Not just more natural conversation—but actually useful for testing ideas and playing devil's advocate. Full episode and detailed blog post linked in the comments 👇 What's your take - will speech-to-speech replace cascaded systems, or will modularity keep cascaded architectures dominant even as naturalness improves?

What if your voice AI could interrupt you the moment it figured out your question - sometimes even before you finished asking it? Last week, I sat down with Neil, CEO of Gradium and co-founder of Kyutai , to talk about the future of speech-to-speech models and why he believes today's cascaded voice systems will soon look "archaic and brittle." Some highlights from our conversation: 🎯 How Kyutai built Moshi—a full duplex conversational AI with "negative latency"—in 6 months with just 4-6 people (while big tech teams had 10-20x the resources) 🧠 Why speech-to-speech models lose intelligence compared to their text counterparts (and what's being done about it) 📱 Pocket TTS: The first voice cloning model that runs on your phone's CPU—not GPU, CPU 🤖 Why robotics and spatial audio represent the next frontier (hint: current voice systems completely break in these environments) 👶 The efficiency gap: Babies learn to speak fluently from <5,000 hours of audio. Current models train on millions of hours. We're doing something wrong. My favorite vision from Neil? The first truly contrarian AI that interrupts you mid-sentence to tell you why you're wrong. Not just more natural conversation—but actually useful for testing ideas and playing devil's advocate. Full episode and detailed blog post linked in the comments 👇 What's your take - will speech-to-speech replace cascaded systems, or will modularity keep cascaded architectures dominant even as naturalness improves?

Brooke Hopkins

13,000 次观看 • 4 个月前

I started learning Plasticity about a year ago, and to train myself, I decided to model 26 models from scratch, one for each letter!

I started learning Plasticity about a year ago, and to train myself, I decided to model 26 models from scratch, one for each letter!

tibo

89,521 次观看 • 2 年前

building large language models from scratch by Sebastian Raschka was a great chance for me to sit down and study again all the LLM basics > token and positional embeddings > self-attention and what QKV is about > causal & multi-head attention studying llms and how they work can seem overwhelming at first. but once you taste how good it feels to learn these things intuitively there's no going back. I shared the resources and my notes on my repo and I hope it's a motivation if you want to start as well or recap.

building large language models from scratch by Sebastian Raschka was a great chance for me to sit down and study again all the LLM basics > token and positional embeddings > self-attention and what QKV is about > causal & multi-head attention studying llms and how they work can seem overwhelming at first. but once you taste how good it feels to learn these things intuitively there's no going back. I shared the resources and my notes on my repo and I hope it's a motivation if you want to start as well or recap.

ℏεsam

49,292 次观看 • 1 年前

So helped me to unlock the voice now it works! It's trained on my Lex Fridman podcast It's 99% perfect but the training data apparently still has some Lex "hummm" sounds in it that sometimes show up I think they call this "crosstalk" and it's hard to remove with AI so I think there's no other choice than to listen to the 1 hour of training data and remove those bits and re-train But then we have a good AI model to turn my book MAKE into an audio book without having to read it for hours/days/weeks

So helped me to unlock the voice now it works! It's trained on my Lex Fridman podcast It's 99% perfect but the training data apparently still has some Lex "hummm" sounds in it that sometimes show up I think they call this "crosstalk" and it's hard to remove with AI so I think there's no other choice than to listen to the 1 hour of training data and remove those bits and re-train But then we have a good AI model to turn my book MAKE into an audio book without having to read it for hours/days/weeks

@levelsio

169,759 次观看 • 1 年前

This is another one of my "Why did I work so hard on this?" Videos. This took 6-8 hours of listening to Sharon Apple yelling to compile and think about everything. I asked myself "How pretentious could I get?" finally winding up at the Vitruvian Man, The Tree Of Life, and Lain.

This is another one of my "Why did I work so hard on this?" Videos. This took 6-8 hours of listening to Sharon Apple yelling to compile and think about everything. I asked myself "How pretentious could I get?" finally winding up at the Vitruvian Man, The Tree Of Life, and Lain.

HoraceVT 🪄🎬

36,217 次观看 • 6 个月前

#WATCH | On L&T Chairman SN Subrahmanyan advocates a 90-hour work week, Mahindra Group Chairman Anand Mahindra says, "I have a huge respect for Narayan Murthy and others. So let me not get this wrong. But I have to say something. I think this debate is in the wrong direction because this debate is about the quantity of work. My point is we have to focus on the quality of work, not on the quantity of work. So it's not about 40 hours, or 70 hours or 90 hours. What output are you doing even if it's 10 hours? You can change the world in 10 hours... I have always believed that you have to have leaders and people in your company who make wise decisions and wise choices. So the question is, which kind of mindmakes the right choices and right decisions? And I think a mind that is exposed to holistic thinking, thinking that is open to inputs from around the world, that's when you'll make a good decision. If you don't spend time at home, if you're not spending time with friends, not reading, and don't have time to reflect, how will you bring the right inputs into making a decision?... I don't want it to be about quantity. Ask me what's the quality of my work. Don't ask me how many hours I work...."

#WATCH | On L&T Chairman SN Subrahmanyan advocates a 90-hour work week, Mahindra Group Chairman Anand Mahindra says, "I have a huge respect for Narayan Murthy and others. So let me not get this wrong. But I have to say something. I think this debate is in the wrong direction because this debate is about the quantity of work. My point is we have to focus on the quality of work, not on the quantity of work. So it's not about 40 hours, or 70 hours or 90 hours. What output are you doing even if it's 10 hours? You can change the world in 10 hours... I have always believed that you have to have leaders and people in your company who make wise decisions and wise choices. So the question is, which kind of mindmakes the right choices and right decisions? And I think a mind that is exposed to holistic thinking, thinking that is open to inputs from around the world, that's when you'll make a good decision. If you don't spend time at home, if you're not spending time with friends, not reading, and don't have time to reflect, how will you bring the right inputs into making a decision?... I don't want it to be about quantity. Ask me what's the quality of my work. Don't ask me how many hours I work...."

ANI

102,689 次观看 • 1 年前

I built an agent that answers machine-learning questions. It's autonomous, and the best part is that I built the whole thing without writing a single line of Python code. Here is what I did and how I did it: Over a year ago, a friend and I built a site that publishes multi-choice questions. You get a new one every day. I decided to have GPT-3.5 answer questions. Here is what I needed to build: 1. Connect to the site's API to retrieve today's question 2. Extract the question and the potential choices 3. Connect to OpenAI's API and ask GPT-3.5 to answer the question 4. Parse the answer from the model 5. Submit the answer back to the API to get the score Not difficult. Likely several hours of work. But I didn't have to write any code. I built the whole thing by dragging and dropping components using Vellum is a YC-backed platform for developers to build LLM applications. They are the only ones I've seen offering this functionality. They sponsored this post, and their team helped me with all my questions while I built this. I created a workflow. The platform supports several node types to build whatever you have in mind. I show how I put the whole thing together in the attached video. The only code I had to write was a few lines of Jinja to parse and transform the API and the LLM results. There are three lessons I want to share from this experience: First, the best possible code is the one you didn't write. I'm a big fan of no-code tools because they help me materialize my ideas fast. They help product people, designers, and no coders collaborate on the solution. Second, Large Language Models are sensitive to how you prompt them. Small changes to prompts can make a big difference in results. This is more pronounced when you are building a multi-step workflow. Third, automated testing and evaluation for prompts is critical. There aren't many companies thinking about this. They'll have a hard time moving from a demo phase. The attached video will show you what I did.

I built an agent that answers machine-learning questions. It's autonomous, and the best part is that I built the whole thing without writing a single line of Python code. Here is what I did and how I did it: Over a year ago, a friend and I built a site that publishes multi-choice questions. You get a new one every day. I decided to have GPT-3.5 answer questions. Here is what I needed to build: 1. Connect to the site's API to retrieve today's question 2. Extract the question and the potential choices 3. Connect to OpenAI's API and ask GPT-3.5 to answer the question 4. Parse the answer from the model 5. Submit the answer back to the API to get the score Not difficult. Likely several hours of work. But I didn't have to write any code. I built the whole thing by dragging and dropping components using Vellum is a YC-backed platform for developers to build LLM applications. They are the only ones I've seen offering this functionality. They sponsored this post, and their team helped me with all my questions while I built this. I created a workflow. The platform supports several node types to build whatever you have in mind. I show how I put the whole thing together in the attached video. The only code I had to write was a few lines of Jinja to parse and transform the API and the LLM results. There are three lessons I want to share from this experience: First, the best possible code is the one you didn't write. I'm a big fan of no-code tools because they help me materialize my ideas fast. They help product people, designers, and no coders collaborate on the solution. Second, Large Language Models are sensitive to how you prompt them. Small changes to prompts can make a big difference in results. This is more pronounced when you are building a multi-step workflow. Third, automated testing and evaluation for prompts is critical. There aren't many companies thinking about this. They'll have a hard time moving from a demo phase. The attached video will show you what I did.

Santiago

309,825 次观看 • 2 年前

New short course: Attention in Transformers: Concepts and Code in PyTorch. Last week we released a course on how LLM transformers work. This week, go deeper and learn about the technical ideas behind the attention mechanism, and see how to code it in PyTorch. This course is built with Joshua Starmer, Founder and CEO of StatQuest. The attention mechanism was a breakthrough that led to transformers, the architecture powering large language models like ChatGPT. Transformers, introduced in the 2017 paper: "Attention is All You Need" by Viswani and others, took off because of its highly scalable design. In this course, you’ll learn how the attention mechanism, a key element of transformer-based LLMs, works and implement it in PyTorch. You'll develop deep intuition about building reliable, functional, and scalable AI applications. What you will do: - Understand the evolution of the attention mechanism, a key breakthrough that led to transformers. - Learn the relationships between word embeddings, positional embeddings, and attention. - Learn about the Query, Key, and Value matrices, and how to produce and use them in attention. - Walk through the math required to calculate self-attention and masked self-attention to learn why and how they work. - Understand the difference between self-attention and masked self-attention and how one is used in the encoder to build context-aware embeddings and the other is used in the decoder for generative outputs. - Learn the details of the encoder-decoder architecture, cross-attention, and multi-head attention and how they are all incorporated into a transformer. - Use PyTorch to code a class that implements self-attention, masked self-attention, and multi-head attention. There're lots of exciting technical details in this course. Please sign up here:

New short course: Attention in Transformers: Concepts and Code in PyTorch. Last week we released a course on how LLM transformers work. This week, go deeper and learn about the technical ideas behind the attention mechanism, and see how to code it in PyTorch. This course is built with Joshua Starmer, Founder and CEO of StatQuest. The attention mechanism was a breakthrough that led to transformers, the architecture powering large language models like ChatGPT. Transformers, introduced in the 2017 paper: "Attention is All You Need" by Viswani and others, took off because of its highly scalable design. In this course, you’ll learn how the attention mechanism, a key element of transformer-based LLMs, works and implement it in PyTorch. You'll develop deep intuition about building reliable, functional, and scalable AI applications. What you will do: - Understand the evolution of the attention mechanism, a key breakthrough that led to transformers. - Learn the relationships between word embeddings, positional embeddings, and attention. - Learn about the Query, Key, and Value matrices, and how to produce and use them in attention. - Walk through the math required to calculate self-attention and masked self-attention to learn why and how they work. - Understand the difference between self-attention and masked self-attention and how one is used in the encoder to build context-aware embeddings and the other is used in the decoder for generative outputs. - Learn the details of the encoder-decoder architecture, cross-attention, and multi-head attention and how they are all incorporated into a transformer. - Use PyTorch to code a class that implements self-attention, masked self-attention, and multi-head attention. There're lots of exciting technical details in this course. Please sign up here:

Andrew Ng

132,216 次观看 • 1 年前

New short course: Open Source Models with Hugging Face 🤗, taught by Maria Khalusova, Marc Sun, and Younes Belkada! Hugging Face has been a game changer by letting you quickly grab any of hundreds of thousands of already-trained open source models to assemble into new applications. This course teaches you best practices for building this way, including how to search and choose among models. You’ll learn to use the Transformers library and walk through multiple models for text, audio, and image processing, including zero-shot image segmentation, zero-shot audio classification, and speech recognition. You'll also learn to use multimodal models for visual question answering, image search, and image captioning. Finally, you’ll learn how to demo what you build locally, on the cloud, or via an API using Gradio and Hugging Face Spaces. You can sign up here:

New short course: Open Source Models with Hugging Face 🤗, taught by Maria Khalusova, Marc Sun, and Younes Belkada! Hugging Face has been a game changer by letting you quickly grab any of hundreds of thousands of already-trained open source models to assemble into new applications. This course teaches you best practices for building this way, including how to search and choose among models. You’ll learn to use the Transformers library and walk through multiple models for text, audio, and image processing, including zero-shot image segmentation, zero-shot audio classification, and speech recognition. You'll also learn to use multimodal models for visual question answering, image search, and image captioning. Finally, you’ll learn how to demo what you build locally, on the cloud, or via an API using Gradio and Hugging Face Spaces. You can sign up here:

Andrew Ng

224,538 次观看 • 2 年前

I trained a 100 million parameter DeepSeek V3 LLM from scratch Here's what you need to know. Previously I trained traditional GPT-2 architecture which has become obsolete with recent LLM advancements. Most recent models like Llama, Mistral, DeepSeek, and GPT-4 use latest architectures. ✦ Model Configuration of my SLM DeepSeek V3 - Parameters: 109,032,032 - Embedding Dimension: 512 - Layers: 8 - Heads: 8 - Experts (MoE): 8 - Experts per token: 2 ✦ DeepSeek brings major architectural changes: - Multi Head Latent Attention - Mixture of Experts - RMS Norm - Multi Token Prediction ✦ Dataset Challenge - TinyStories is great for learning SLMs. I trained GPT-2 on it previously with good results. - But I needed a more challenging dataset. - If I use TinyStories again on DeepSeek, how would I know MHLA, MoE or MTP works better than old architecture? - The old architecture can handle it, so new DeepSeek would too without utilizing latest advancements. That's why I moved to FineWeb-Edu dataset Thanks Yuvraj Singh (smolhub.com) for the suggestion for this dataset ✦ Training Journey - Rented A100 PCIe GPU and trained the model. - Did test runs. During final run, model was 65% trained but stopped due to glitch after 4 hours. - Fixed all edge cases and ran training again with increased config parameters. - Final training: 7 hours, 20,000 epochs 𝐓𝐨𝐭𝐚𝐥 𝐆𝐏𝐔 𝐜𝐨𝐬𝐭: $17 - $9.53 for main 7-hour run - $7.42 for experiments and demos ✦ Reflection Amazing long project that taught me latest architectural advancements. I'll reimplement and revisit after a few weeks because there's too much complexity, mostly in Multi Head Latent Attention part. Need to make concepts stronger. Code Final trained Model Dataset Resources Huge shoutout to Raj Dandekar again for creating one of the most detailed video series about DeepSeek - this was my primary resource for the implementation. Playlist Blogs by Maarten Grootendorst These are excellent visual blogs to understand MoE in detail. Thanks Maarten for your amazing contributions to the community through your books and blogs Blogs on MoE Implemention of MoE from scratch by @aviTwit3 One of the most detailed blogs on implementing Mixture of Experts. Thanks Avinash for this blog - it helped me understand Mixture of Experts much better. If you're someone in the 𝐌𝐋 & 𝐋𝐋𝐌 space, would love to 𝐜𝐨𝐧𝐧𝐞𝐜𝐭 and discuss this field in general, so give a follow up for that.

I trained a 100 million parameter DeepSeek V3 LLM from scratch Here's what you need to know. Previously I trained traditional GPT-2 architecture which has become obsolete with recent LLM advancements. Most recent models like Llama, Mistral, DeepSeek, and GPT-4 use latest architectures. ✦ Model Configuration of my SLM DeepSeek V3 - Parameters: 109,032,032 - Embedding Dimension: 512 - Layers: 8 - Heads: 8 - Experts (MoE): 8 - Experts per token: 2 ✦ DeepSeek brings major architectural changes: - Multi Head Latent Attention - Mixture of Experts - RMS Norm - Multi Token Prediction ✦ Dataset Challenge - TinyStories is great for learning SLMs. I trained GPT-2 on it previously with good results. - But I needed a more challenging dataset. - If I use TinyStories again on DeepSeek, how would I know MHLA, MoE or MTP works better than old architecture? - The old architecture can handle it, so new DeepSeek would too without utilizing latest advancements. That's why I moved to FineWeb-Edu dataset Thanks Yuvraj Singh (smolhub.com) for the suggestion for this dataset ✦ Training Journey - Rented A100 PCIe GPU and trained the model. - Did test runs. During final run, model was 65% trained but stopped due to glitch after 4 hours. - Fixed all edge cases and ran training again with increased config parameters. - Final training: 7 hours, 20,000 epochs 𝐓𝐨𝐭𝐚𝐥 𝐆𝐏𝐔 𝐜𝐨𝐬𝐭: $17 - $9.53 for main 7-hour run - $7.42 for experiments and demos ✦ Reflection Amazing long project that taught me latest architectural advancements. I'll reimplement and revisit after a few weeks because there's too much complexity, mostly in Multi Head Latent Attention part. Need to make concepts stronger. Code Final trained Model Dataset Resources Huge shoutout to Raj Dandekar again for creating one of the most detailed video series about DeepSeek - this was my primary resource for the implementation. Playlist Blogs by Maarten Grootendorst These are excellent visual blogs to understand MoE in detail. Thanks Maarten for your amazing contributions to the community through your books and blogs Blogs on MoE Implemention of MoE from scratch by @aviTwit3 One of the most detailed blogs on implementing Mixture of Experts. Thanks Avinash for this blog - it helped me understand Mixture of Experts much better. If you're someone in the 𝐌𝐋 & 𝐋𝐋𝐌 space, would love to 𝐜𝐨𝐧𝐧𝐞𝐜𝐭 and discuss this field in general, so give a follow up for that.

Mayank Pratap Singh

48,005 次观看 • 1 年前

This is the best and fastest speech-to-text model in the world: • 23.2 seconds to process 30 minutes of audio • 93.3% accuracy • Diarization support to detect multiple speakers • Trained on 12.5 million hours of multilingual data I tried it out and it's pretty impressive:

This is the best and fastest speech-to-text model in the world: • 23.2 seconds to process 30 minutes of audio • 93.3% accuracy • Diarization support to detect multiple speakers • Trained on 12.5 million hours of multilingual data I tried it out and it's pretty impressive:

Santiago

66,397 次观看 • 10 个月前

New short course: Building Multimodal Search and RAG", by Weaviate AI Database's Sebastia(N_) Witalec ✊🏽✊🏾✊🏿. Contrastive learning is used to train models to map vectors into an embedding space by pulling similar concepts closer together and pushing dissimilar concepts away from each other. This technique is also used to train multimodal embedding models that capture semantic similarity across different modalities like text, images, and audio. These multimodal embeddings can be used to build multimodal search and RAG systems. In this course, you'll learn how contrastive learning works, and how to add multimodality to RAG – so your models can draw on diverse, relevant context to answer questions. For example, a query about a financial report might synthesize information from text snippets, graphs, tables, and slides. You will also learn how visual instruction tuning lets you integrate image understanding into language models, and build a multi-vector recommender system using Weaviate’s open-source vector database. Please sign up here:

New short course: Building Multimodal Search and RAG", by Weaviate AI Database's Sebastia(N_) Witalec ✊🏽✊🏾✊🏿. Contrastive learning is used to train models to map vectors into an embedding space by pulling similar concepts closer together and pushing dissimilar concepts away from each other. This technique is also used to train multimodal embedding models that capture semantic similarity across different modalities like text, images, and audio. These multimodal embeddings can be used to build multimodal search and RAG systems. In this course, you'll learn how contrastive learning works, and how to add multimodality to RAG – so your models can draw on diverse, relevant context to answer questions. For example, a query about a financial report might synthesize information from text snippets, graphs, tables, and slides. You will also learn how visual instruction tuning lets you integrate image understanding into language models, and build a multi-vector recommender system using Weaviate’s open-source vector database. Please sign up here:

Andrew Ng

104,371 次观看 • 2 年前

Text-to-speech is moving way too fast. Just a few days ago, I tweeted about PersonaPlex-7B, NVIDIA's new open source TTS ( And today, Qwen just open-sourced Qwen3-TTS 🤯 It’s a revolutionary text-to-speech model built for control. Not just about generating speech, but about shaping how it sounds directly from language. You can guide the pace, the tone, and the expressiveness straight from text, without touching audio graphs or hand-tuning parameters. That’s the real shift! What makes Qwen3-TTS stand out is how practical it already is: → voice cloning from just a few seconds of audio → voice creation without any reference sample → support for 10 languages out of the box → end-to-end latency down to ~97ms → works in both streaming and non-streaming setups The models come in two sizes (0.6B and 1.7B), so you can trade off quality and hardware cost depending on your setup. You can work with curated voices, designed voices, or cloned ones, and it integrates cleanly with vLLM for production use. It also ships as a simple Python package you can pip install. If you’re building real-time voice systems, this removes a lot of friction! 100% free and open source. I put the repo in the 🧵↓

Text-to-speech is moving way too fast. Just a few days ago, I tweeted about PersonaPlex-7B, NVIDIA's new open source TTS ( And today, Qwen just open-sourced Qwen3-TTS 🤯 It’s a revolutionary text-to-speech model built for control. Not just about generating speech, but about shaping how it sounds directly from language. You can guide the pace, the tone, and the expressiveness straight from text, without touching audio graphs or hand-tuning parameters. That’s the real shift! What makes Qwen3-TTS stand out is how practical it already is: → voice cloning from just a few seconds of audio → voice creation without any reference sample → support for 10 languages out of the box → end-to-end latency down to ~97ms → works in both streaming and non-streaming setups The models come in two sizes (0.6B and 1.7B), so you can trade off quality and hardware cost depending on your setup. You can work with curated voices, designed voices, or cloned ones, and it integrates cleanly with vLLM for production use. It also ships as a simple Python package you can pip install. If you’re building real-time voice systems, this removes a lot of friction! 100% free and open source. I put the repo in the 🧵↓

Charly Wargnier

59,144 次观看 • 5 个月前

Fitness Update: End of 3 months of Bulking. Starting a 3 month cut tomorrow, my last cut of 2026. I lost about 50 pounds from January 2025 till the end of March 2026 with a 13 month straight cut aided by Retatrutide and Tirzepatide. Counted calories and ran a deficit and lifted weights the whole time. My body needed a break and a reset from the stress of cutting for that long and constant downregulating of my metabolism. I decided to take April off and just do more calories and more lifting, and that turned into deciding to do May and June as well. So for 3 months straight I've been eating a lot more and lifting a LOT. I've gained about 5 pounds. Abs are not very visible anymore. But I've put on some muscle and strength as well and increased my base metabolic rate by several hundred calories. This is positioning me to be able to better be able to run a deficit again and continue into another cut with more energy and success possible. So this is me thicc boi mode ready to start a cut and get lean and mean. I'm sitting at ~195 right now and ~16-18% body fat. Goal of my cut is to get to the 180 or so range over the next 3 months and be in the 12-13% BF range, at which point I'll do a slow patient bulk through Oct-January and enjoy the fall and Holidays with lots of calories and strength training. Starting tomorrow I'm going to add 30-45 minutes of daily fasted zone 2 cardio first thing in the morning to burn an extra 150-300 cals. And I will drop my daily calorie allowance about 400 calories from what I've been eating the last few months. (2500-2800 down to 2200 to start, lowering later to 2100 and 2000 lowest if necessary to keep rate of loss where I want) This should get me the 1+ pound per week rate of loss I'm looking for. And I'll continue to prioritize lifting and protein as well to preserve the muscle mass I've built up and get as much of the loss to be fat as possible. I'm also going zero alcohol the whole cut as well. If I execute well, I'll be by far the fittest and leanest I've ever been at the end of September and I'm excited to see what that looks and feels like when I get there. Going from 70+ pounds overweight to building your dream physique isn't a quick process, people. Give yourself a couple of years to work on it systematically. It took you years to get out of shape, so you can take a couple years to get into great shape. But once you're there it's easier to stay there than it was to get there thankfully. The work continues. Let me know if you have any questions about anything. I'm documenting the whole process and will keep y'all up to speed. Open book. Building in public for accountability and hopefully to help others with their journey as well. 4th pic is me at my fattest in 2024. Embarrassing to post but necessary conext. I was in a sad, sad state you guys and heading towards diabetes and an early grave. If I can do this, you can do this!

Fitness Update: End of 3 months of Bulking. Starting a 3 month cut tomorrow, my last cut of 2026. I lost about 50 pounds from January 2025 till the end of March 2026 with a 13 month straight cut aided by Retatrutide and Tirzepatide. Counted calories and ran a deficit and lifted weights the whole time. My body needed a break and a reset from the stress of cutting for that long and constant downregulating of my metabolism. I decided to take April off and just do more calories and more lifting, and that turned into deciding to do May and June as well. So for 3 months straight I've been eating a lot more and lifting a LOT. I've gained about 5 pounds. Abs are not very visible anymore. But I've put on some muscle and strength as well and increased my base metabolic rate by several hundred calories. This is positioning me to be able to better be able to run a deficit again and continue into another cut with more energy and success possible. So this is me thicc boi mode ready to start a cut and get lean and mean. I'm sitting at ~195 right now and ~16-18% body fat. Goal of my cut is to get to the 180 or so range over the next 3 months and be in the 12-13% BF range, at which point I'll do a slow patient bulk through Oct-January and enjoy the fall and Holidays with lots of calories and strength training. Starting tomorrow I'm going to add 30-45 minutes of daily fasted zone 2 cardio first thing in the morning to burn an extra 150-300 cals. And I will drop my daily calorie allowance about 400 calories from what I've been eating the last few months. (2500-2800 down to 2200 to start, lowering later to 2100 and 2000 lowest if necessary to keep rate of loss where I want) This should get me the 1+ pound per week rate of loss I'm looking for. And I'll continue to prioritize lifting and protein as well to preserve the muscle mass I've built up and get as much of the loss to be fat as possible. I'm also going zero alcohol the whole cut as well. If I execute well, I'll be by far the fittest and leanest I've ever been at the end of September and I'm excited to see what that looks and feels like when I get there. Going from 70+ pounds overweight to building your dream physique isn't a quick process, people. Give yourself a couple of years to work on it systematically. It took you years to get out of shape, so you can take a couple years to get into great shape. But once you're there it's easier to stay there than it was to get there thankfully. The work continues. Let me know if you have any questions about anything. I'm documenting the whole process and will keep y'all up to speed. Open book. Building in public for accountability and hopefully to help others with their journey as well. 4th pic is me at my fattest in 2024. Embarrassing to post but necessary conext. I was in a sad, sad state you guys and heading towards diabetes and an early grave. If I can do this, you can do this!

Clint Fiore 🦬 DM for Biz Deals

75,761 次观看 • 14 天前

I’m non-technical and built the app I always wanted with AI. I like art and history, and wanted an easy way to learn more about both. I take a photo, wait a few seconds, and get a two-minute audio tour of what I’m seeing. Feels wild to go from non-technical to vibe coding an entire app in 30 days.

I’m non-technical and built the app I always wanted with AI. I like art and history, and wanted an easy way to learn more about both. I take a photo, wait a few seconds, and get a two-minute audio tour of what I’m seeing. Feels wild to go from non-technical to vibe coding an entire app in 30 days.

Marta Jamrozik

254,615 次观看 • 1 年前

It’s hard to believe that 3 years ago, none of this was possible. Now, it’s pretty much going to change everything. Last month, Fish Audio reached out to me to try their voice AI software. They just launched their new S1 model today, and I was curious about the state of AI and voice (and everything else), so I gave it a spin. I liked it enough that when I was asked about a partnership, I said yes. I’m free to talk about what I like and don’t like about the software, and I’m going to share with you a few voices I cloned as well as some of the workflow. It was much easier than I thought. To give you an example, here is a private (research only) voice I cloned as a test. I grabbed a clip of Rutger Hauer’s famous speech from Blade Runner and uploaded it to the voice cloner on the Fish Audio Website as a private voice (no one else can use it, as it is for research only). I didn’t think it would work. The audio sample is very short. But Fish Audio was able to clone the voice extremely well and very fast. I didn’t have to upload any more than that to produce these results. I used Grok to write the new dialog, and I added the rain and background effects and the result is pretty impressive. It only took just a few minutes once I had the audio uploaded for everything from cloning to generating multiple takes. I’ll give some tips and pointers on how to get the best results at the conclusion of this thread and show you some surprising things it can do. (con’t) #Promotion

It’s hard to believe that 3 years ago, none of this was possible. Now, it’s pretty much going to change everything. Last month, Fish Audio reached out to me to try their voice AI software. They just launched their new S1 model today, and I was curious about the state of AI and voice (and everything else), so I gave it a spin. I liked it enough that when I was asked about a partnership, I said yes. I’m free to talk about what I like and don’t like about the software, and I’m going to share with you a few voices I cloned as well as some of the workflow. It was much easier than I thought. To give you an example, here is a private (research only) voice I cloned as a test. I grabbed a clip of Rutger Hauer’s famous speech from Blade Runner and uploaded it to the voice cloner on the Fish Audio Website as a private voice (no one else can use it, as it is for research only). I didn’t think it would work. The audio sample is very short. But Fish Audio was able to clone the voice extremely well and very fast. I didn’t have to upload any more than that to produce these results. I used Grok to write the new dialog, and I added the rain and background effects and the result is pretty impressive. It only took just a few minutes once I had the audio uploaded for everything from cloning to generating multiple takes. I’ll give some tips and pointers on how to get the best results at the conclusion of this thread and show you some surprising things it can do. (con’t) #Promotion

Grummz

69,859 次观看 • 1 年前

I just hit 100k subscribers on YouTube and I'm stoked. Which is why I've decided to give this whole content creation thing a real shot. Starting today, I am going to stream consistently on Twitch and regularly upload to my channel. A little back story on how I got here. Before I created content I suffered from clinical depression after having my thyroid removed due to cancer. I would escape by watching streams and gaming videos to relax. One day I decided to just give it a shot. So I hooked up my computer and talked into a very cheap mic and started streaming. It went terribly, but honestly... I was thrilled. Then I took some of that footage and edited it down like I did with skate videos to some random song that reminded me of a model runway. Took forever to get done on my free time. I got one encouraging comment... And I was absolutely thrilled. These days I get comments from people who tell me what they're struggling through and how my videos helped them forget about it for an hour. And every time that happens, I'm reminded of why I do this. To pay back for all the times I was entertained when I was having a hard time. It seems so dumb, but man... I can't tell you what a stupid little video or silly stream can do for someone having a really hard time. I'm not sure if this is going to work... But I'm going to give it a try. Even if I fail, I'm not going to stop creating. I got exciting plans for the future but this is the first step. Where it takes me... I'm not sure. I hope you all come along. Regardless. Thanks to all of you for supporting me and helping me out. I never thought I'd hit 100k or have any type of following. Not in a million years. I am grateful and appreciative. Here's to what comes next I guess. Tea is life. ☕️

I just hit 100k subscribers on YouTube and I'm stoked. Which is why I've decided to give this whole content creation thing a real shot. Starting today, I am going to stream consistently on Twitch and regularly upload to my channel. A little back story on how I got here. Before I created content I suffered from clinical depression after having my thyroid removed due to cancer. I would escape by watching streams and gaming videos to relax. One day I decided to just give it a shot. So I hooked up my computer and talked into a very cheap mic and started streaming. It went terribly, but honestly... I was thrilled. Then I took some of that footage and edited it down like I did with skate videos to some random song that reminded me of a model runway. Took forever to get done on my free time. I got one encouraging comment... And I was absolutely thrilled. These days I get comments from people who tell me what they're struggling through and how my videos helped them forget about it for an hour. And every time that happens, I'm reminded of why I do this. To pay back for all the times I was entertained when I was having a hard time. It seems so dumb, but man... I can't tell you what a stupid little video or silly stream can do for someone having a really hard time. I'm not sure if this is going to work... But I'm going to give it a try. Even if I fail, I'm not going to stop creating. I got exciting plans for the future but this is the first step. Where it takes me... I'm not sure. I hope you all come along. Regardless. Thanks to all of you for supporting me and helping me out. I never thought I'd hit 100k or have any type of following. Not in a million years. I am grateful and appreciative. Here's to what comes next I guess. Tea is life. ☕️

TeaGuyTom

21,510 次观看 • 1 年前

I trained a 12M parameter LLM on my own ML framework using a Rust backend and CUDA kernels for flash attention, AdamW, and more. Wrote the full transformer architecture, and BPE tokenizer from scratch. The framework features: - Custom CUDA kernels (Flash Attention, fused LayerNorm, fused GELU) for 3x increased throughput - Automatic WebGPU fallback for non-NVIDIA devices - TypeScript API with Rust compute backend - One npm install to get started, prebuilt binaries for every platform Try out the model for yourself: Built with Reese Chong. Check out the repos and blog if you want to learn more. Shoutout to Modal for the compute credits allowing me to train on 2 A100 GPUs without going broke cc sunny madra Gavin

I trained a 12M parameter LLM on my own ML framework using a Rust backend and CUDA kernels for flash attention, AdamW, and more. Wrote the full transformer architecture, and BPE tokenizer from scratch. The framework features: - Custom CUDA kernels (Flash Attention, fused LayerNorm, fused GELU) for 3x increased throughput - Automatic WebGPU fallback for non-NVIDIA devices - TypeScript API with Rust compute backend - One npm install to get started, prebuilt binaries for every platform Try out the model for yourself: Built with Reese Chong. Check out the repos and blog if you want to learn more. Shoutout to Modal for the compute credits allowing me to train on 2 A100 GPUs without going broke cc sunny madra Gavin

Aadi Kulshrestha

812,043 次观看 • 2 个月前

Introducing Parrot 🦜 Our first speech-to-text model built for production-grade voice agents with an industry leading accuracy, built for noisy calls, regional dialects and compressed audio. Trained on thousands of hours of real-world audio to make sure when you say Udaipur, the agent doesn't hear 'you-die-poor'.

Introducing Parrot 🦜 Our first speech-to-text model built for production-grade voice agents with an industry leading accuracy, built for noisy calls, regional dialects and compressed audio. Trained on thousands of hours of real-world audio to make sure when you say Udaipur, the agent doesn't hear 'you-die-poor'.

Vedant

15,714 次观看 • 1 个月前