mirror of
https://github.com/jambonz/batch-speech-utils.git
synced 2026-05-06 08:16:57 +00:00
restructure
This commit is contained in:
@@ -1,7 +1,14 @@
|
||||
const { getTranscription } = require('./lib/transcribe');
|
||||
const { getRedactedAudio } = require('./lib/redact');
|
||||
|
||||
module.exports = {
|
||||
getTranscription,
|
||||
getRedactedAudio
|
||||
const noopLogger = {
|
||||
info: () => {},
|
||||
debug: () => {},
|
||||
error: () => {}
|
||||
};
|
||||
|
||||
module.exports = (logger) => {
|
||||
logger = logger || noopLogger;
|
||||
|
||||
return {
|
||||
transcribe: require('./lib/transcribe').bind(null, logger),
|
||||
redact: require('./lib/redact').bind(null, logger),
|
||||
};
|
||||
};
|
||||
|
||||
@@ -0,0 +1,3 @@
|
||||
module.exports = {
|
||||
transcribe: require('./transcribe')
|
||||
};
|
||||
@@ -0,0 +1,13 @@
|
||||
const fs = require('fs');
|
||||
const { createClient } = require('@deepgram/sdk');
|
||||
|
||||
const transcribe = async(logger, apiKey, filePath) => {
|
||||
logger.info(`Transcribing audio file: ${filePath}`);
|
||||
const client = createClient(apiKey);
|
||||
const fileBuffer = fs.readFileSync(filePath);
|
||||
const options = { model: 'nova-2', smart_format: true, detect_entities: true };
|
||||
const { result } = await client.listen.prerecorded.transcribeFile(fileBuffer, options);
|
||||
return result;
|
||||
};
|
||||
|
||||
module.exports = transcribe;
|
||||
@@ -1,59 +0,0 @@
|
||||
const fs = require('fs');
|
||||
|
||||
async function transcribeFile(deepgram, filePath) {
|
||||
const fileBuffer = fs.readFileSync(filePath);
|
||||
const options = { model: 'nova-2', smart_format: true, detect_entities: true };
|
||||
const { result } = await deepgram.listen.prerecorded.transcribeFile(fileBuffer, options);
|
||||
return result;
|
||||
}
|
||||
|
||||
async function redactFile(deepgram, filePath) {
|
||||
const fileBuffer = fs.readFileSync(filePath);
|
||||
const options = { model: 'nova-2', smart_format: true, redact: 'pii' };
|
||||
const { result } = await deepgram.listen.prerecorded.transcribeFile(fileBuffer, options);
|
||||
return result;
|
||||
}
|
||||
|
||||
async function analyzeText(deepgram, text) {
|
||||
const options = { language: 'en', sentiment: true, intents: true, summarize: true };
|
||||
const { result } = await deepgram.read.analyzeText({ text }, options);
|
||||
return result;
|
||||
}
|
||||
|
||||
|
||||
async function processAudio(deepgram, filePath) {
|
||||
try {
|
||||
const transcription = await transcribeFile(deepgram, filePath);
|
||||
const redaction = await redactFile(deepgram, filePath);
|
||||
const transcript = transcription.results.channels[0].alternatives[0].transcript;
|
||||
const timestamps = transcription.results.channels[0].alternatives[0].words;
|
||||
const redactionTimestamps = redaction.results.channels[0].alternatives[0].words;
|
||||
const redacted = redaction.results.channels[0].alternatives[0].transcript;
|
||||
const entities = transcription.results.channels[0].alternatives[0].entities;
|
||||
|
||||
const analysisResult = await analyzeText(deepgram, transcript);
|
||||
const sentimentSegment = analysisResult.results.sentiments.segments[0];
|
||||
const sentiment = sentimentSegment.sentiment;
|
||||
const sentimentScore = sentimentSegment.sentiment_score;
|
||||
|
||||
return {
|
||||
transcript,
|
||||
timestamps,
|
||||
redactionTimestamps,
|
||||
redacted,
|
||||
sentiment,
|
||||
sentimentScore,
|
||||
entities
|
||||
};
|
||||
} catch (error) {
|
||||
console.error('Error processing audio:', error);
|
||||
throw error;
|
||||
}
|
||||
}
|
||||
|
||||
module.exports = {
|
||||
transcribeFile,
|
||||
analyzeText,
|
||||
processAudio,
|
||||
redactFile
|
||||
};
|
||||
@@ -1,46 +0,0 @@
|
||||
const ffmpeg = require('fluent-ffmpeg');
|
||||
|
||||
|
||||
async function redactAudioDeepgram(transcriptionData, audioPath, audioOutputPath, { delta = 0.05 } = {}) {
|
||||
return new Promise((resolve, reject) => {
|
||||
const command = ffmpeg(audioPath)
|
||||
.outputFormat('wav'); // Ensure output format is WAV
|
||||
|
||||
// Iterate over transcription data to apply audio filters
|
||||
transcriptionData.forEach((data, i) => {
|
||||
const { word, start } = data;
|
||||
let end = data.end; // Default end time
|
||||
|
||||
// Check if the word needs redaction
|
||||
if (word.startsWith('[') && word.endsWith(']')) {
|
||||
// Find the start of the next non-redacted word
|
||||
for (let j = i + 1; j < transcriptionData.length; j++) {
|
||||
if (!(transcriptionData[j].word.startsWith('[') && transcriptionData[j].word.endsWith(']'))) {
|
||||
end = transcriptionData[j].start;
|
||||
break;
|
||||
}
|
||||
}
|
||||
command.audioFilters({
|
||||
filter: 'volume',
|
||||
options: `volume=0:enable='between(t,${start - delta},${end})'` // Applying silence
|
||||
});
|
||||
|
||||
// Log the redacted segments
|
||||
console.log(`Redacting from ${start}s to ${end}s: "${word}"`);
|
||||
}
|
||||
});
|
||||
|
||||
// Handlers for command execution
|
||||
command.on('end', () => {
|
||||
console.log(`Redacted audio saved at ${audioOutputPath}`);
|
||||
resolve(); // Resolve the promise on successful completion
|
||||
}).on('error', (err, stdout, stderr) => {
|
||||
console.error('Error processing audio file:', err.message);
|
||||
console.error('ffmpeg stdout:', stdout);
|
||||
console.error('ffmpeg stderr:', stderr);
|
||||
reject(err); // Reject the promise on error
|
||||
}).saveToFile(audioOutputPath);
|
||||
});
|
||||
}
|
||||
|
||||
module.exports = { redactAudioDeepgram };
|
||||
+42
-12
@@ -1,16 +1,46 @@
|
||||
const { redactAudioDeepgram } = require('./make-redact-audio-deepgram');
|
||||
const ffmpeg = require('fluent-ffmpeg');
|
||||
|
||||
const assert = require('assert');
|
||||
const redact = async(logger, {vendor, transcriptionData, audioPath, audioOutputPath, delta = 0.05}) => {
|
||||
logger.info(`Redacting audio file: ${audioPath} vendor: ${vendor}`);
|
||||
return new Promise((resolve, reject) => {
|
||||
const command = ffmpeg(audioPath)
|
||||
.outputFormat('wav'); // Ensure output format is WAV
|
||||
|
||||
function getRedactedAudio(vendor, transcriptionData, audioPath, audioOutputPath) {
|
||||
assert.ok(['deepgram', 'otherVendor'].includes(vendor), 'vendor not supported');
|
||||
// Iterate over transcription data to apply audio filters
|
||||
transcriptionData.forEach((data, i) => {
|
||||
const { word, start } = data;
|
||||
let end = data.end; // Default end time
|
||||
|
||||
if (vendor === 'deepgram') {
|
||||
return redactAudioDeepgram(transcriptionData, audioPath, audioOutputPath);
|
||||
}
|
||||
else {
|
||||
throw new Error(`Unsupported vendor: ${vendor}`);
|
||||
}
|
||||
}
|
||||
// Check if the word needs redaction
|
||||
if (word.startsWith('[') && word.endsWith(']')) {
|
||||
// Find the start of the next non-redacted word
|
||||
for (let j = i + 1; j < transcriptionData.length; j++) {
|
||||
if (!(transcriptionData[j].word.startsWith('[') && transcriptionData[j].word.endsWith(']'))) {
|
||||
end = transcriptionData[j].start;
|
||||
break;
|
||||
}
|
||||
}
|
||||
command.audioFilters({
|
||||
filter: 'volume',
|
||||
options: `volume=0:enable='between(t,${start - delta},${end})'` // Applying silence
|
||||
});
|
||||
|
||||
module.exports = { getRedactedAudio };
|
||||
// Log the redacted segments
|
||||
console.log(`Redacting from ${start}s to ${end}s: "${word}"`);
|
||||
}
|
||||
});
|
||||
|
||||
// Handlers for command execution
|
||||
command.on('end', () => {
|
||||
console.log(`Redacted audio saved at ${audioOutputPath}`);
|
||||
resolve(); // Resolve the promise on successful completion
|
||||
}).on('error', (err, stdout, stderr) => {
|
||||
console.error('Error processing audio file:', err.message);
|
||||
console.error('ffmpeg stdout:', stdout);
|
||||
console.error('ffmpeg stderr:', stderr);
|
||||
reject(err); // Reject the promise on error
|
||||
}).saveToFile(audioOutputPath);
|
||||
});
|
||||
};
|
||||
|
||||
module.exports = redact;
|
||||
|
||||
+12
-15
@@ -1,19 +1,16 @@
|
||||
const { createDeepGramClient } = require('./utils');
|
||||
const fs = require('fs');
|
||||
const { processAudio } = require('./get-transcription-deepgram');
|
||||
const assert = require('assert');
|
||||
const {transcribe:dgTranscribe} = require('./deepgram');
|
||||
|
||||
function getTranscription(vendor, apiKey, filePath) {
|
||||
assert.ok(['deepgram'].includes(vendor), 'vendor not supported');
|
||||
if (vendor === 'deepgram') {
|
||||
const deepgramClient = createDeepGramClient(apiKey);
|
||||
assert.ok(deepgramClient, 'Invalid Deepgram API key');
|
||||
return processAudio(deepgramClient, filePath);
|
||||
}
|
||||
else {
|
||||
throw new Error(`Unsupported vendor: ${vendor}`);
|
||||
const transcribe = async(logger, credentials, filePath) => {
|
||||
const { vendor } = credentials;
|
||||
|
||||
switch (credentials.vendor) {
|
||||
case 'deepgram':
|
||||
assert.ok(credentials.apiKey, 'Deepgram API key is required');
|
||||
return await dgTranscribe(logger, credentials.apiKey, filePath);
|
||||
default:
|
||||
throw new Error(`Unsupported vendor: ${vendor}`);
|
||||
}
|
||||
};
|
||||
|
||||
}
|
||||
|
||||
module.exports = {getTranscription};
|
||||
module.exports = transcribe;
|
||||
|
||||
@@ -1,9 +0,0 @@
|
||||
|
||||
const { createClient } = require('@deepgram/sdk');
|
||||
|
||||
function createDeepGramClient(DEEPGRAM_API_KEY) {
|
||||
return createClient(DEEPGRAM_API_KEY);
|
||||
}
|
||||
|
||||
|
||||
module.exports = {createDeepGramClient};
|
||||
Reference in New Issue
Block a user