tflite & scorer & audio files

[i2rllc]

This is extremely novice but what directory do the tflite, scorer and audio files get stored to to execute STT? I put it in coqui-stt-venv and coqui-stt-venv/bin and it's not executing so clearly the files are in the wrong directory; please help. TYIA

[HarikalarKutusu]

What are you trying to do? What is your environment?

In any case, you can put your data anywhere and point to them with command-line arguments in training for example.

You can find the related arguments by using --help with each command and they are explained here:
https://stt.readthedocs.io/en/latest/TRAINING_FLAGS.html

For example, in the notebook, which ran on Google Colab, I kept the data on Google Drive and expand it to the Colab instance (Ubuntu) somewhere and pointed to them during calling the related commands:
https://github.com/HarikalarKutusu/common-voice-tr-experiments/blob/main/colab-notebooks/tr/v8.0/r4/stt-train-cv-tr-v8.0-r4-s5.ipynb

Or for inference, for the Voice Chess application, I used a server and put it data somewhere, and point to them:
https://github.com/HarikalarKutusu/3d-voice-chess/blob/897ab6598460b5a5cae10f9f69fc1f10c6426efa/voice-chess-server/server.js#L29

[i2rllc]

@wasertech Thank you!

1. Why use gist or dpaste? What does it do that screenshot doesn't?
2. This is probably going to be very stupid question but how do I send the output to a file. I tried executing the command followed by > filename.txt but it didn't actually save anything? I must be missing something. Thank you again

[wasertech]

Why use gist or dpaste? What does it do that screenshot doesn't?

You want our help? You share your logs the proper way. It's that simple. We need to be able to copy and paste your logs, search trough them, etc.

This is probably going to be very stupid question but how do I send the output to a file. I tried executing the command followed by > filename.txt but it didn't actually save anything

Try like this:

❯ stt --model output_graph.tflite --scorer kenlm.scorer  --audio audio.wav > transcript.txt
Loading model from file output_graph.tflite
TensorFlow: v2.8.0-8-g06c8fea58fd
 Coqui STT: v1.3.0-0-g148fa743
INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
Loaded model in 0.00897s.
Loading scorer from files kenlm.scorer
Loaded scorer in 0.0037s.
Running inference.
Inference took 1.075s for 3.580s audio file.
❯ cat transcript.txt
───────┬───────────────────────────────────────────────────────────────────────────
       │ File: transcript.txt
───────┼───────────────────────────────────────────────────────────────────────────
   1   │ bonjour
───────┴───────────────────────────────────────────────────────────────────────────

You can also checkout listen which allows to directly transcribe into a file like so:

❯ listen -f audio.wav
Filename                       Duration(s)         
audio.wav                     3.580               

❯ cat audio.txt
Bonjour.

Read more

[i2rllc]

@wasertech Thank you. I am new to this so your effort to teach me is appreciated.

[HarikalarKutusu]

Why use gist or dpaste? What does it do that screenshot doesn't?

For example, I saw your post yesterday night, but I was on mobile. I could not see the details in the image, so I could not answer.

[i2rllc]

@HarikalarKutusu understood, thank you