I'm looking for a software (for Windows or Linux) that can convert MP3/WAV files to text.
Something that can generate automatically text with speech recognition, like the automatic subtitles on YouTube, but that could output these subtitles in a text file.
This is the online demo of IBM Watson Speech to Text service. You may also use with any of SDKs available on their page. It can take a WAV file but not MP3, so you will want to convert MP3s first.
On the page, select the language you want to use, and whatever or not you want to try to identify multiple speakers. Upload the audio file and wait for the result.
It is by no means perfect. Due to its nature, it lacks the chance to train it to recognize a particular voice well, which is an advantage that transcription software have (but those often require microphone input, of course, you could try an audio loop back). On the other hand, Watson is constantly improving and has the capability to recognize multiple speakers (in some cases) which is something transcription software cannot do (and does not intend to do).
On Windows there's the Speech Recognition API, which is very easy to use, as in a few function calls, and you're all set.
https://msdn.microsoft.com/en-us/library/jj127860.aspx
Otherwise, albeit slower, you can use Google Cloud Speech API, or IBM Watson, and Amazon's Alexa Voice Server to do the same with internet connection.
You can always redirect audio output into audio capture device (you can do that on windows with the audio loopback device, and most OSes support a way of doing this), in which case all the audio output of your computer, such as YouTube audio, can be captured by the said APIs.
Unfortunately I don't specifically know a dedicated application for this, but for Windows this might contain what you need.
https://www.microsoft.com/en-us/download/details.aspx?id=27224
Otherwise a bit of programming skill, it should be easy to do it — I would be surprise if no one hasn't yet.
An extensive list of speech recognition software, some of which may contain the ability to save text output to a file I found here:
https://en.wikipedia.org/wiki/Speech_recognition_software_for_Linux
and
https://en.wikipedia.org/wiki/List_of_speech_recognition_software
Using Office 365 you have 300 minutes of speech recognition per month available directly in Word:
OpenAI's Whisper (off-line, MIT license, Python 3.9, CLI) yields some highly accurate transcription. To use (tested on Ubuntu 20.04 x64 LTS but it also works on Windows):
conda create -y --name whisperpy39 python==3.9
conda activate whisperpy39
pip install git+https://github.com/openai/whisper.git
sudo apt update && sudo apt install ffmpeg
whisper recording.wav
whisper recording.wav --model large
If using an Nvidia 3090 GPU, add the following after conda activate whisperpy39
pip install -f https://download.pytorch.org/whl/torch_stable.html
conda install pytorch==1.10.1 torchvision torchaudio cudatoolkit=11.0 -c pytorch
Performance info below.
Model inference time:
| Size | Parameters | English-only model | Multilingual model | Required VRAM | Relative speed |
|---|---|---|---|---|---|
| tiny | 39 M | tiny.en |
tiny |
~1 GB | ~32x |
| base | 74 M | base.en |
base |
~1 GB | ~16x |
| small | 244 M | small.en |
small |
~2 GB | ~6x |
| medium | 769 M | medium.en |
medium |
~5 GB | ~2x |
| large | 1550 M | N/A | large |
~10 GB | 1x |
WER on several corpus from https://cdn.openai.com/papers/whisper.pdf:
WER on several languages from https://github.com/openai/whisper/blob/main/language-breakdown.svg:
This is an intriguing question. Full disclosure: I don't have expertise in this.
As discussed in the related question: Windows tool to set voice as password, what you are seeking may be more properly called speech recognition, not voice recognition. I did a tag search for "speech recognition" and got 24 hits (Speech Recognition search). One in particular may be of interest to you, Speech Recognition software for windows that takes audio file as input and outputs text file. Its answers cite two programs, Dragon Naturally Speaking, which is commercial and probably not of interest, and CMUSpinx, which is open source. The answer indicates you may need to do some programming and that your mp3 would have to be converted to a wav file.
Let us know if this works out!
You may want to try "Dragon NaturallySpeaking", this an expensive software which is advertised to do what you want. I had the opportunity to test an older version and I would say that voice recognition still has quite a way to go before become actually useful for end user.
