CloudTranscriptionProvider.recordAndTranscribe() is blocking — needs async refactor and service integration #1

New issue

Closed

opened 2026-05-12 16:50:11 +00:00 by Mark · 1 comment

Mark commented

2026-05-12 16:50:11 +00:00

Owner

Problem

CloudTranscriptionProvider.recordAndTranscribe() uses Thread.sleep() for chunk timing and OkHttp.newCall(request).execute() (blocking) for HTTP requests. It is designed to be called from a coroutine loop, but:

Thread.sleep() blocks the calling thread — should use delay() in a suspend function
OkHttp.execute() blocks — should use enqueue() or kotlinx.coroutines with withContext(Dispatchers.IO)
TranscriptionService.kt doesn't use the TranscriptionProvider abstraction at all — it still hardcodes android.speech.SpeechRecognizer directly. The cloud backend is never actually activated in the service loop, so this entire path is dead code in production.

Location

app/src/main/java/com/voicetranscriber/service/CloudTranscriptionProvider.kt (lines 75–80, 102, 150)
app/src/main/java/com/voicetranscriber/service/TranscriptionService.kt (lines 93–167)

Suggested approach

Make recordAndTranscribe() a suspend function, replace Thread.sleep() with kotlinx.coroutines.delay(), and use withContext(Dispatchers.IO) for the blocking HTTP call
Refactor TranscriptionService.kt to use SpeechBackend.select() to pick a backend at startup, then run the appropriate continuous loop:
- For ANDROID/FIREOS: existing SpeechRecognizer callback pattern
- For CLOUD: coroutine loop calling recordAndTranscribe() and feeding results through the listener
Store and read BackendConfig from SharedPreferences at service startup
Consider suspendCancellableCoroutine for the OkHttp call if keeping the current HTTP client

Part of closing the gap between the provider abstraction layer and the actual service implementation.

## Problem `CloudTranscriptionProvider.recordAndTranscribe()` uses `Thread.sleep()` for chunk timing and `OkHttp.newCall(request).execute()` (blocking) for HTTP requests. It is designed to be called from a coroutine loop, but: 1. `Thread.sleep()` blocks the calling thread — should use `delay()` in a `suspend` function 2. `OkHttp.execute()` blocks — should use `enqueue()` or `kotlinx.coroutines` with `withContext(Dispatchers.IO)` 3. **`TranscriptionService.kt` doesn't use the `TranscriptionProvider` abstraction at all** — it still hardcodes `android.speech.SpeechRecognizer` directly. The cloud backend is never actually activated in the service loop, so this entire path is dead code in production. ## Location - `app/src/main/java/com/voicetranscriber/service/CloudTranscriptionProvider.kt` (lines 75–80, 102, 150) - `app/src/main/java/com/voicetranscriber/service/TranscriptionService.kt` (lines 93–167) ## Suggested approach 1. Make `recordAndTranscribe()` a `suspend` function, replace `Thread.sleep()` with `kotlinx.coroutines.delay()`, and use `withContext(Dispatchers.IO)` for the blocking HTTP call 2. Refactor `TranscriptionService.kt` to use `SpeechBackend.select()` to pick a backend at startup, then run the appropriate continuous loop: - For `ANDROID`/`FIREOS`: existing `SpeechRecognizer` callback pattern - For `CLOUD`: coroutine loop calling `recordAndTranscribe()` and feeding results through the listener 3. Store and read `BackendConfig` from SharedPreferences at service startup 4. Consider `suspendCancellableCoroutine` for the OkHttp call if keeping the current HTTP client Part of closing the gap between the provider abstraction layer and the actual service implementation.