Audio-to-Text is a Google Cloud Function that accepts an audio file and sends it to the Google Cloud Speech API for translation. The Cloud function is triggered by posting a file to an HTTP endpoint.
We've used this on a React Native app built with Expo. Expo records an audio file, which is sent to the Google Function.
The pipeline is built on two components:
- Google Cloud Functions
- Google Cloud Speech API
You will need to sign up for Google Cloud services (including billing), create a project and enable the Speech API. To deploy the function from local, you will need to install the Google Cloud SDK. More on that here.
One of the tricky parts about this is to match the config to the file you are trying to translate. On iOS I have had the best luck with .wav files at 41000 Hz.
This Cloud Function includes a hardcoded config:
- Format: LINEAR16
- Frequency: 41000 Hz (16000 or higher is recommended)
- Channels: Mono (a Cloud Speech API requirement)
- Language: English (US)
Included in this repo is a test audio file that will work if you change the hZ to 24000.
I also used ffmpeg to get the encoding, hZ, etc for an audio file:
brew install ffmpeg
ffmpeg -i ~/Documents/speech-test.wav -hide_banner
To create the function and deploy it:
gcloud functions deploy audioToText --trigger-http --runtime nodejs8
After it's created:
gcloud functions deploy audioToText
- https://cloud.google.com/functions/docs/writing/http - Multipart Data
- Helpful example of audio to text pipeline: https://github.com/jlaham/audio-2-text