ben/AndyTranscribe
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Use local faster-whisper only for transcription.

Browse Source 
Drop cloud and Ollama engine choices so audio stays on-machine via Docker Whisper, and tighten orphan detection plus UI around a single local flow.
This commit is contained in:
ben
2026-08-12 15:34:16 +02:00
parent bc6cb5efa4
commit 21b17c7657
10 changed files with 117 additions and 295 deletions
Download Patch File Download Diff File Expand all files Collapse all files
app/Services/TranscriptionService.php
+1 -99
View File
@@ -4,52 +4,18 @@ namespace App\Services;
use App\Models\Recording;
use Closure;
use Illuminate\Support\Facades\Http;
use Illuminate\Support\Str;
use Laravel\Ai\Transcription;
use RuntimeException;
class TranscriptionService
{
/**
* Run transcription for a recording using the selected driver.
* Transcribe a recording with the local faster-whisper server.
*
* @param (Closure(string, int): void)|null $onProgress
*/
public function transcribe(Recording $recording, ?Closure $onProgress = null): string
{
$report = $onProgress ?? static fn (string $message, int $percent) => null;
return match ($recording->transcription_driver) {
'cloud' => $this->viaCloud($recording, $report),
'local' => $this->viaLocal($recording, $report),
'ollama' => $this->viaRemoteCompatible($recording, $report),
default => throw new RuntimeException('Unknown transcription driver: '.$recording->transcription_driver),
};
}
/**
* @param Closure(string, int): void $report
*/
private function viaCloud(Recording $recording, Closure $report): string
{
$report('Sending audio to OpenAI Whisper…', 35);
$report('Waiting for cloud transcript (this can take a while for long recordings)…', 55);
$transcript = Transcription::fromStorage($recording->file_path)
->timeout((int) config('ai.transcription_timeout', 600))
->generate('openai', 'whisper-1');
$report('Received transcript from OpenAI…', 85);
return (string) $transcript;
}
/**
* @param Closure(string, int): void $report
*/
private function viaLocal(Recording $recording, Closure $report): string
{
$model = config('ai.local_whisper_model', 'Systran/faster-whisper-base');
$report('Connecting to local faster-whisper server…', 30);
@@ -63,68 +29,4 @@ class TranscriptionService
return (string) $transcript;
}
/**
* Call an OpenAI-compatible /v1/audio/transcriptions endpoint at a user-supplied host URL.
*
* @param Closure(string, int): void $report
*/
private function viaRemoteCompatible(Recording $recording, Closure $report): string
{
if (! filled($recording->ollama_url)) {
throw new RuntimeException('Ollama host URL is required for remote transcription.');
}
$base = $this->normalizeBaseUrl($recording->ollama_url);
$model = config('ai.remote_whisper_model', config('ai.local_whisper_model', 'Systran/faster-whisper-base'));
$path = $recording->absolutePath();
if (! is_readable($path)) {
throw new RuntimeException('Recording audio file is not readable.');
}
$report('Connecting to remote host '.$recording->ollama_url.'…', 30);
$report("Uploading audio and waiting for transcript ({$model})…", 50);
$response = Http::timeout((int) config('ai.transcription_timeout', 600))
->attach(
'file',
fopen($path, 'r'),
$recording->original_filename ?: basename($path),
)
->post($base.'/audio/transcriptions', [
'model' => $model,
'response_format' => 'json',
]);
if (! $response->successful()) {
throw new RuntimeException(
'Remote transcription failed (HTTP '.$response->status().'): '.$response->body()
);
}
$text = $response->json('text');
if (! is_string($text) || $text === '') {
throw new RuntimeException('Remote transcription returned an empty transcript. Ensure the host exposes OpenAI-compatible /v1/audio/transcriptions.');
}
$report('Received transcript from remote host…', 85);
return $text;
}
/**
* Normalize a user URL to an OpenAI-style base ending in /v1.
*/
private function normalizeBaseUrl(string $url): string
{
$url = rtrim(trim($url), '/');
if (Str::endsWith($url, '/v1')) {
return $url;
}
return $url.'/v1';
}
}