ben.carrasco/batch-speech-utils
1
0
Fork 0
mirror of https://github.com/jambonz/batch-speech-utils.git synced 2026-07-24 13:12:21 +00:00
Code Issues Packages Projects Releases Wiki Activity

updating for transcription schema

Browse Source
This commit is contained in:
surajshivakumar
2024-06-26 19:51:18 -04:00
parent 4a6cc284a7
commit 90ecd734b4
5 changed files with 189 additions and 156 deletions
Download Patch File Download Diff File Expand all files Collapse all files
lib/deepgram/config.js
+4 -2
View File
@@ -1,13 +1,15 @@
const transcriptionOptions = {
model: 'nova-2',
smart_format: true,
detect_entities: true
detect_entities: true,
multichannel:true
};
const redactionOptions = {
model: 'nova-2',
smart_format: true,
redact: 'pii'
redact: 'pii',
multichannel:true
};
const analysisOptions = {
lib/deepgram/transcribe.js
+49 -23
View File
@@ -2,6 +2,39 @@ const fs = require('fs');
const { createClient } = require('@deepgram/sdk');
const { transcriptionOptions, redactionOptions, analysisOptions } = require('./config');
function extractTranscript(data) {
// eslint-disable-next-line max-len
const paragraphs = data.results.channels.flatMap((channel) => channel.alternatives.flatMap((alt) => alt.paragraphs.paragraphs));
let ctr = 0;
// Use reduce to process each paragraph and sentence, consolidating transcripts by speaker
return paragraphs.reduce((acc, paragraph) => {
paragraph.sentences.forEach((sentence) => {
const wordsDetails = data.results.channels
.find((channel) => channel.alternatives.some((alt) => alt.paragraphs.paragraphs.includes(paragraph)))
.alternatives[0].words
.filter((word) => word.start >= sentence.start && word.end <= sentence.end)
.map((word) => ({
word: word.word,
start: word.start,
end: word.end,
confidence: word.confidence
}));
acc.push({
timestamp: sentence.start,
duration: Math.round(1000 * (sentence.end - sentence.start)),
startTime: sentence.start,
endTime: sentence.end,
speaker: ctr++ % 2,
transcript: sentence.text,
words: wordsDetails
});
});
return acc;
}, []);
}
const transcribe = async(logger, apiKey, filePath) => {
logger.info(`Transcribing audio file: ${filePath}`);
//creating a deepgram client
@@ -9,32 +42,25 @@ const transcribe = async(logger, apiKey, filePath) => {
//audio file buffer
const fileBuffer = fs.readFileSync(filePath);
//transcription
const { result:transcriptResult } = await client.listen.prerecorded.transcribeFile(fileBuffer, transcriptionOptions);
const transcript = transcriptResult.results.channels[0].alternatives[0].transcript;
const timestamps = transcriptResult.results.channels[0].alternatives[0].words;
const entities = transcriptResult.results.channels[0].alternatives[0].entities;
const confidence = transcriptResult.results.channels[0].alternatives[0].confidence;
// const { result:transcriptResult } = await client.listen.prerecorded.transcribeFile(fileBuffer, transcriptionOptions);
//redaction
const { result:redactionResult } = await client.listen.prerecorded.transcribeFile(fileBuffer, redactionOptions);
const redactionTimestamps = redactionResult.results.channels[0].alternatives[0].words;
const redacted = redactionResult.results.channels[0].alternatives[0].transcript;
//analysis and sentiment
const { result:analysisResult } = await client.read.analyzeText({ text:transcript }, analysisOptions);
const sentimentSegment = analysisResult.results.sentiments.segments[0];
const sentiment = sentimentSegment.sentiment;
const sentimentScore = sentimentSegment.sentiment_score;
const vendor = 'deepgram';
return {
vendor,
transcript,
timestamps,
redactionTimestamps,
redacted,
sentiment,
sentimentScore,
entities,
confidence
const data = {
'vendor' : 'deepgram',
'model' : redactionResult.metadata.model_info[redactionResult.metadata.models[0]].arch,
'channels' : redactionResult.metadata.channels,
'createdAt': redactionResult.metadata.created
};
data.speechEvents = extractTranscript(redactionResult);
const combinedTranscript = data.speechEvents.map(event => event.transcript).join(" ");
data.redactionTimestamps = data.speechEvents.flatMap(event => event.words);
//analysis and sentiment
const { result:analysisResult } = await client.read.analyzeText({ text:combinedTranscript }, analysisOptions);
const sentimentSegment = analysisResult.results.sentiments.segments[0];
data.sentiment = sentimentSegment.sentiment;
data.sentimentScore = sentimentSegment.sentiment_score;
data.totalDuration = Math.round(1000 * redactionResult.metadata.duration);
return data;
};
module.exports = transcribe;