Mozilla Releases DeepSpeech 0.7 As Their Great Speech-To-Text Engine

[pal666]

without rewrite of tensorflow in rust this announcement feels incomplete

[jf33]

By the way, you can contribute to Common Voice not only in English, but also in many other languages. Perfect activity for those being caught in quarantine.

[oleid]

By the way, this is not CUDA only, it also works fine with AMD's ROCm. Easy to set up with their tensorflow docker container. It also might work with a recent upstream kernel, however I had to use their dkms driver to get my R470 going.

[pabloski]

FYI: Mycroft is also using DeepSpeech. Still waiting for my Mark II though.

This project has always fascinated me. A true, offline, opensource, digital assistant, without all the bullcrap from Amazon, Google and (spy)friends.

[pabloski]

It is pretty scary what you can do with machine learning. With a 5 second audio clip you feed that to a machine learning software and clone that voice and make it say whatever you want. So you can make it sound like the president, or your friend, or if you have an enemy, it can make it sound like your enemy. Someone could make it sound like you and and make it say that it admits to murdering someone.

Same with photomontage. And like the only technologies, this new tech leaves artifacts, that forensic expert are able to identify to say if it say a fake or true voice. And the game of cat and mouse goes on and on.

[stfn]

FYI: Mycroft is also using DeepSpeech. Still waiting for my Mark II though.

[Okki]

It is not a voice recognition engine. It is a text-to-speech engine.

The Mozilla text-to-speech project is TTS.

[gripped]

It is not a voice recognition engine. It is a text-to-speech engine.

Best open an issue on their github and tell them they've got it the wrong way round in the readme.md

[uid313]

It is pretty scary what you can do with machine learning. With a 5 second audio clip you feed that to a machine learning software and clone that voice and make it say whatever you want. So you can make it sound like the president, or your friend, or if you have an enemy, it can make it sound like your enemy. Someone could make it sound like you and and make it say that it admits to murdering someone.

[Okki]

I take this opportunity to recall the existence of the Mozilla Common Voice project, to which everyone can contribute, to be able to provide free data to the DeepSpeech engine, and thus build a free speech recognition software, as well as a quality speech synthesis.