Kyutai Speech to Text 1B & 2.6B Local Setup

In this video we'll be testing Kyutai's speech-to-text models locally with Python. Checking out both the 1B English/French model and 2.6B English-only model using the terminal and a Gradio web app, to see if their streaming/realtime transcription capabilities actually work as advertised.

00:00 Intro: Kyutai STT Model
00:09 Kyutai TTS Demo
00:14 Kyutai STT Setup Details
00:30 Kyutai STT Realtime Transcription Demo
01:21 Kyutai's Github Repo
02:14 Kyutai STT Model Variants
03:03 Local Setup Begins
09:09 First Test: Realtime Transcription in the Terminal (MLX Model)
10:43 Kyutai STT Pytorch Model Setup
16:25 Second Test: Realtime Transcription with the Pytorch Model
17:08 Gradio Web UI Overview
19:42 Third Test: Realtime Transcription in Gradio Web App
20:40 Fourth Test: Multilingual Transcription Test
21:55 Fifth Test: Kyutai 2.6B Model Multilingual Test
23:05 Sixth Test: Kyutai STT 2.6B Realtime Transcription Gradio Web UI
24:01 Seventh Test: Longer Audio File Transcription
26:48 Audio Fiile Transcription Issue
29:25 Final Test
30:25 Final Remarks
31:40 Outro

🔗 LINKS
HF Repo:
1B Model: https://huggingface.co/kyutai/stt-1b-...
2.6B Model: https://huggingface.co/kyutai/stt-2.6...

Official Github Repo: https://github.com/kyutai-labs/delaye...
Project Github Repo: https://github.com/brainiakk/kyutai

#kyutai #stt #speechtotext #localstt #offlinestt #realtimetranscription #streamingstt #gradio #gradioui #cpuinference #ondevice #opensource #voiceai #speechrecognition #modelsetup #moshi #moshimlx #huggingface #tutorial #demo #developer #privacy #lowlatency #transcriptiondemo #speechmodels #whisperalternative #installguide #runlocally #microphoneinput #livetranscript #edgeai #edgeinference #kyutaistt