kobotoolbox / kpi / 25379279264 / 3

82%

master: 76%

push

github

feat(nlp): upgrade automatic Google transcription to Speech-to-Text v2 batch API DEV-1498 (#6955)

### 📣 Summary
This PR upgrades the automatic Google transcription service from the
Speech-to-Text v1 API to the v2 API.

The implementation uploads audio files to Google Cloud Storage, submits
a batch recognition job, and retrieves the resulting transcripts from
JSON output files written by Google.

The service now dynamically resolves the correct region and model using
the new languages configuration introduced in the languages app.

### 📖 Description
The new implementation follows the v2 architecture:
1. Audio files are uploaded to Google Cloud Storage
2. A BatchRecognize job is created
3. Google processes the audio asynchronously
4. Transcription results are written as JSON files to Cloud Storage
5. The service reads the output JSON and returns the transcript

The implementation supports two mechanisms for tracking long-running
operations:
- Cache-based tracking: Used for single transcription requests.
- Database-based tracking: Future bulk transcription operations will
store the Google operation id on `SubsequenceBulkActionItem`.

5150 of 12081 branches covered (42.63%)

20094 of 36402 relevant lines covered (55.2%)

0.55 hits per line