Improve Transcribe error diagnostics, add minimum audio check
- Log $response status/headers/body on deserialization errors - Add 10ms delay between audio chunks to prevent stream overload - Skip transcription if audio < 0.5s (too short for recognition) - Cleaner error logging with dedicated logTranscribeError helper
This commit is contained in:
Daniel Onyejesi
parent
42b002eea8
commit
a36cd9a299
1 changed files with 47 additions and 27 deletions
|
|
@ -61,7 +61,23 @@ function convertToPCM(inputBuffer) {
|
|||
async function* makeAudioStream(buffer) {
|
||||
for (var i = 0; i < buffer.length; i += CHUNK_SIZE) {
|
||||
yield { AudioEvent: { AudioChunk: buffer.slice(i, i + CHUNK_SIZE) } };
|
||||
await new Promise(function(r) { setTimeout(r, 0); });
|
||||
// Small delay between chunks to prevent overwhelming the stream
|
||||
await new Promise(function(r) { setTimeout(r, 10); });
|
||||
}
|
||||
}
|
||||
|
||||
function logTranscribeError(label, err) {
|
||||
console.error('[Transcribe] ' + label + ':', err.message);
|
||||
if (err.name) console.error('[Transcribe] name:', err.name);
|
||||
if (err.$metadata) console.error('[Transcribe] metadata:', JSON.stringify(err.$metadata));
|
||||
if (err.cause) console.error('[Transcribe] cause:', err.cause.message || JSON.stringify(err.cause));
|
||||
if (err.$response) {
|
||||
try {
|
||||
var resp = err.$response;
|
||||
console.error('[Transcribe] status:', resp.statusCode);
|
||||
if (resp.headers) console.error('[Transcribe] headers:', JSON.stringify(resp.headers));
|
||||
if (resp.body) console.error('[Transcribe] body:', typeof resp.body === 'string' ? resp.body.slice(0, 500) : '(stream)');
|
||||
} catch (e) { /* ignore logging errors */ }
|
||||
}
|
||||
}
|
||||
|
||||
|
|
@ -75,8 +91,6 @@ async function transcribeWithAWS(audioBuffer, mimeType) {
|
|||
secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY
|
||||
} : undefined;
|
||||
|
||||
var client = new TranscribeStreamingClient({ region: region, credentials: credentials });
|
||||
|
||||
// Try ffmpeg conversion to PCM (most reliable with AWS Transcribe).
|
||||
// Falls back to ogg-opus passthrough if ffmpeg is not installed.
|
||||
var audioToSend = audioBuffer;
|
||||
|
|
@ -90,13 +104,18 @@ async function transcribeWithAWS(audioBuffer, mimeType) {
|
|||
console.log('[Transcribe] ffmpeg conversion OK — PCM 16kHz, ' + audioToSend.length + ' bytes');
|
||||
} catch (ffErr) {
|
||||
console.warn('[Transcribe] ffmpeg unavailable, sending ogg-opus directly:', ffErr.message);
|
||||
// For WAV input, try pcm anyway without ffmpeg
|
||||
if (mimeType && mimeType.indexOf('wav') !== -1) {
|
||||
mediaEncoding = 'pcm';
|
||||
sampleRate = 16000;
|
||||
}
|
||||
}
|
||||
|
||||
// Minimum ~0.5s of PCM audio at 16kHz mono 16-bit = 16000 bytes
|
||||
if (mediaEncoding === 'pcm' && audioToSend.length < 16000) {
|
||||
console.warn('[Transcribe] Audio too short (' + audioToSend.length + ' bytes), skipping');
|
||||
return '';
|
||||
}
|
||||
|
||||
var useMedical = process.env.AWS_TRANSCRIBE_MEDICAL === 'true';
|
||||
var specialty = process.env.AWS_TRANSCRIBE_SPECIALTY || 'PRIMARYCARE';
|
||||
var transcript = '';
|
||||
|
|
@ -118,10 +137,11 @@ async function transcribeWithAWS(audioBuffer, mimeType) {
|
|||
return text;
|
||||
}
|
||||
|
||||
// Try Medical first if enabled, with fallback to Standard
|
||||
if (useMedical) {
|
||||
try {
|
||||
var StartMedicalStreamTranscriptionCommand = TranscribeModule.StartMedicalStreamTranscriptionCommand;
|
||||
var medCmd = new StartMedicalStreamTranscriptionCommand({
|
||||
var client = new TranscribeStreamingClient({ region: region, credentials: credentials });
|
||||
var medCmd = new TranscribeModule.StartMedicalStreamTranscriptionCommand({
|
||||
LanguageCode: 'en-US',
|
||||
MediaSampleRateHertz: sampleRate,
|
||||
MediaEncoding: mediaEncoding,
|
||||
|
|
@ -131,29 +151,29 @@ async function transcribeWithAWS(audioBuffer, mimeType) {
|
|||
});
|
||||
var medResp = await client.send(medCmd);
|
||||
transcript = await collectTranscript(medResp.TranscriptResultStream);
|
||||
console.log('[Transcribe] Medical transcription OK (' + transcript.length + ' chars)');
|
||||
console.log('[Transcribe] Medical OK (' + transcript.length + ' chars)');
|
||||
} catch (medErr) {
|
||||
console.warn('[Transcribe] Medical failed:', medErr.message);
|
||||
if (medErr.name) console.warn('[Transcribe] Error name:', medErr.name);
|
||||
if (medErr.$metadata) console.warn('[Transcribe] Metadata:', JSON.stringify(medErr.$metadata));
|
||||
if (medErr.cause) console.warn('[Transcribe] Cause:', medErr.cause.message || medErr.cause);
|
||||
console.warn('[Transcribe] Falling back to standard Transcribe...');
|
||||
// Fallback to standard Transcribe
|
||||
var client2 = new TranscribeStreamingClient({ region: region, credentials: credentials });
|
||||
var StartStreamTranscriptionCommand = TranscribeModule.StartStreamTranscriptionCommand;
|
||||
var cmd = new StartStreamTranscriptionCommand({
|
||||
LanguageCode: 'en-US',
|
||||
MediaSampleRateHertz: sampleRate,
|
||||
MediaEncoding: mediaEncoding,
|
||||
AudioStream: makeAudioStream(audioToSend)
|
||||
});
|
||||
var resp = await client2.send(cmd);
|
||||
transcript = await collectTranscript(resp.TranscriptResultStream);
|
||||
console.log('[Transcribe] Standard fallback OK (' + transcript.length + ' chars)');
|
||||
logTranscribeError('Medical failed', medErr);
|
||||
console.warn('[Transcribe] Falling back to standard...');
|
||||
try {
|
||||
var client2 = new TranscribeStreamingClient({ region: region, credentials: credentials });
|
||||
var stdCmd = new TranscribeModule.StartStreamTranscriptionCommand({
|
||||
LanguageCode: 'en-US',
|
||||
MediaSampleRateHertz: sampleRate,
|
||||
MediaEncoding: mediaEncoding,
|
||||
AudioStream: makeAudioStream(audioToSend)
|
||||
});
|
||||
var stdResp = await client2.send(stdCmd);
|
||||
transcript = await collectTranscript(stdResp.TranscriptResultStream);
|
||||
console.log('[Transcribe] Standard fallback OK (' + transcript.length + ' chars)');
|
||||
} catch (stdErr) {
|
||||
logTranscribeError('Standard also failed', stdErr);
|
||||
throw stdErr;
|
||||
}
|
||||
}
|
||||
} else {
|
||||
var StartStreamTranscriptionCommand = TranscribeModule.StartStreamTranscriptionCommand;
|
||||
var cmd = new StartStreamTranscriptionCommand({
|
||||
var client = new TranscribeStreamingClient({ region: region, credentials: credentials });
|
||||
var cmd = new TranscribeModule.StartStreamTranscriptionCommand({
|
||||
LanguageCode: 'en-US',
|
||||
MediaSampleRateHertz: sampleRate,
|
||||
MediaEncoding: mediaEncoding,
|
||||
|
|
@ -161,7 +181,7 @@ async function transcribeWithAWS(audioBuffer, mimeType) {
|
|||
});
|
||||
var resp = await client.send(cmd);
|
||||
transcript = await collectTranscript(resp.TranscriptResultStream);
|
||||
console.log('[Transcribe] Standard transcription OK (' + transcript.length + ' chars)');
|
||||
console.log('[Transcribe] Standard OK (' + transcript.length + ' chars)');
|
||||
}
|
||||
|
||||
return transcript.trim();
|
||||
|
|
|
|||
Loading…
Reference in a new issue