Larynx Demo

Channel:synesthesiam · Length:6:39 · Published:2021-08-24 · Views:13822

Watch on YouTube

Summary

A demonstration of Larynx, an offline open source neural text-to-speech system supporting nine languages. The video itself is narrated with a Larynx-generated voice.

Why HN recommends it

HN points to it as proof that offline TTS can sound natural, relevant to private voice assistant alternatives.

Larynx is significantly ahead in terms of quality of output — HN, 2021

Computer vision, speech & NLP

Subtitles (excerpt)

[Music] welcome to a demonstration of the larynx text-to-speech system larynx runs completely offline even on a raspberry pi the open source libraries are used to transform text into phonemes and voices trained on open datasets produced speech the result isn't perfect but is hopefully good enough to avoid the need for a cloud service to get started you can install larynx using docker a debian package or inside a python virtual environment the docker installation is easiest so this is what we'll be using for demonstration the larynx server script runs an http server inside a docker container on port 5002 [Music] [Music] this is a test new voices may be downloaded from within the web interface which will be stored in your local directory [Music] [Music] this is a test notice that the first sentence spoken by each voice takes a bit longer because it has to be loaded first this is a test on a raspberry pi 4 the low quality setting must be used to get faster than real-time performance this is a test an http engine point is available for synthesizing text externally [Music] there's an end point for that [Music] of all the advanced settings the denoiser strength is the most important [Music] setting this value to low can

On Hacker News

Found this project the other day that is probably good enough quality for most uses i.e. sounds very natural and not like a robot from the 1980s and doesn't require a cloud service and can run on modest hardware [LINK] Demo: [VIDEO] — sublinear, 2023-02
It's unfortunate that another comment that links to Rhasspy has been downvoted (I assume because it lacked any other context) so I wanted to mention the project with some additional context: [LINK] While I've not used the entire Rhasspy project myself (but trying it out is on the long list of things to do :) ) I have used the offline Text-To-Speech sub-project Larynx... ...and it is amazing ! Lar… — follower, 2021-12
yeah, [VIDEO] at the end says southern female english [LINK] but the sample is NOT like the video, maybe the samples are old. the video is a great example — 2Gkashmiri, 2021-12
So, after actually watching the demo video all the way through, it seems the video is narrated by a Larynx-produced voice "southern_english_female": [VIDEO] While I appreciate the cinematic aspect of the confirmation/"reveal" at the conclusion of the video, ironically--because the quality is so good--since most people won't get through a 7 minute video, it would be entirely possible to not realis… — follower, 2021-08

Related videos