Bad inference when using live transcript

Asma_Trabelsi · June 4, 2021, 12:34pm

Hello,

I try to use deep speech live transcription. I have used this tutorial

mozilla/DeepSpeech-examples/blob/r0.9/web_microphone_websocket/Readme.md

# Web Microphone Websocket

This is an example of a ReactJS web application streaming microphone audio from the browser
to a NodeJS server and transmitting the DeepSpeech results back to the browser.

#### Download the pre-trained model (1.8GB):

```
wget https://github.com/mozilla/DeepSpeech/releases/download/v0.9.3/deepspeech-0.9.3-models.pbmm
wget https://github.com/mozilla/DeepSpeech/releases/download/v0.9.3/deepspeech-0.9.3-models.scorer
```

#### Install:

```
yarn install
```

#### Run ReactJS Client:

This file has been truncated. show original

The transcription seems to be very bad!!!

Any help please to ameliorate quality ?

dkreutz · June 4, 2021, 10:17pm

Try recording some audio with the same microphone setup and listen to it. Is there any loud noise, electrical hum?
You can try to use another microphone, speak louder and closer to the microphone and try to reduce background noise (computer fan, air condition, …).

Do speakers have a strong accent, speak very fast or is voice attenuated (ffp2 mask)?
Try speaking slower and with clear intonation.