{"slug": "how-ai-learned-to-speak-a-rudrite-research-film-rudrite-research", "title": "How AI Learned to Speak — a Rudrite Research film | Rudrite Research", "summary": "Rudrite Research released a 47-minute documentary explaining how modern text-to-speech works, from sound waves as numbers to neural audio codecs and talking LLMs. The film covers key milestones like WaveNet, Tacotron, VALL-E, and current systems, aiming to make TTS technology accessible to a broad audience.", "body_md": "# How AI Learned to Speak\n\nA from-zero ramp through how modern text-to-speech actually works: a sound wave as numbers, the neural codec that turns audio into tokens, TTS as next-token prediction, all the way to today’s talking LLMs — built on the verified explainers behind it.\n\n47 min · 27 chapters · [Watch on YouTube ↗](https://www.youtube.com/watch?v=8W_qiZqocwo)\n\n## Chapters\n\n- 0:00 — TTS as next-token prediction\n- 0:50 — Sound is just numbers\n- 2:47 — The mel spectrogram\n- 4:13 — From text to phonemes\n- 5:37 — The alignment problem\n- 6:12 — Before deep learning\n- 6:56 — WaveNet\n- 7:43 — Tacotron and attention\n- 9:46 — FastSpeech and the variance adaptor\n- 12:15 — Flow, diffusion, and VITS\n- 13:38 — Audio as tokens: VQ and RVQ\n- 15:51 — Neural audio codecs\n- 19:09 — VALL-E and the three-slot frame\n- 21:37 — A gallery of modern systems\n- 24:58 — How these models are trained\n- 28:42 — Post-training with GRPO\n- 29:41 — Inference and the latency problem\n- 31:29 — Making it fast\n- 33:48 — Controlling prosody, voice, and emotion\n- 36:13 — A decade in one sentence\n- 36:57 — Moshi and Whisper\n- 38:12 — How we evaluate TTS\n- 39:17 — The whole field on one map\n- 40:13 — Reading a 2026 paper\n- 41:02 — The gist\n- 41:53 — Bonus: FACodec, AudioLM, and Tortoise\n- 44:08 — Bonus: multilingual, safety, and beyond", "url": "https://wpnews.pro/news/how-ai-learned-to-speak-a-rudrite-research-film-rudrite-research", "canonical_source": "https://research.rudrite.com/videos/how-ai-learned-to-speak", "published_at": "2026-06-26 00:00:00+00:00", "updated_at": "2026-06-27 07:08:22.761955+00:00", "lang": "en", "topics": ["machine-learning", "large-language-models", "generative-ai", "ai-research", "ai-products"], "entities": ["Rudrite Research", "WaveNet", "Tacotron", "VALL-E", "FastSpeech", "VITS", "Moshi", "Whisper"], "alternates": {"html": "https://wpnews.pro/news/how-ai-learned-to-speak-a-rudrite-research-film-rudrite-research", "markdown": "https://wpnews.pro/news/how-ai-learned-to-speak-a-rudrite-research-film-rudrite-research.md", "text": "https://wpnews.pro/news/how-ai-learned-to-speak-a-rudrite-research-film-rudrite-research.txt", "jsonld": "https://wpnews.pro/news/how-ai-learned-to-speak-a-rudrite-research-film-rudrite-research.jsonld"}}