feat: simple coqui-ai stt integration

package.json

@@ -78,6 +78,7 @@
     "node-wav": "0.0.2",
     "socket.io": "^4.4.0",
     "socket.io-client": "^4.4.0",
+    "stt": "^1.1.0",
     "superagent": "^6.1.0"
   },
   "devDependencies": {

scripts/setup-offline/setup-stt.js

@@ -10,23 +10,23 @@ import os from '@/helpers/os'
 export default () => new Promise(async (resolve, reject) => {
   log.info('Setting up offline speech-to-text...')
 
-  const destDeepSpeechFolder = 'bin/deepspeech'
+  const destCoquiFolder = 'bin/coqui'
   const tmpDir = 'scripts/tmp'
-  const deepSpeechVersion = '0.9.3'
+  const coquiVersion = '1.0.0'
   let downloader = 'wget'
   if (os.get().type === 'macos') {
     downloader = 'curl -L -O'
   }
 
-  if (!fs.existsSync(`${destDeepSpeechFolder}/deepspeech.scorer`)) {
+  if (!fs.existsSync(`${destCoquiFolder}/model.tflite`)) {
     try {
       log.info('Downloading pre-trained model...')
-      await command(`cd ${tmpDir} && ${downloader} https://github.com/mozilla/DeepSpeech/releases/download/v${deepSpeechVersion}/deepspeech-${deepSpeechVersion}-models.pbmm`, { shell: true })
-      await command(`cd ${tmpDir} && ${downloader} https://github.com/mozilla/DeepSpeech/releases/download/v${deepSpeechVersion}/deepspeech-${deepSpeechVersion}-models.scorer`, { shell: true })
+      await command(`cd ${tmpDir} && ${downloader} https://github.com/coqui-ai/STT-models/releases/download/english/coqui/v${coquiVersion}-huge-vocab/model.tflite`, { shell: true })
+      await command(`cd ${tmpDir} && ${downloader} https://github.com/coqui-ai/STT-models/releases/download/english/coqui/v${coquiVersion}-huge-vocab/huge-vocabulary.scorer`, { shell: true })
       log.success('Pre-trained model download done')
       log.info('Moving...')
-      await command(`mv -f ${tmpDir}/deepspeech-${deepSpeechVersion}-models.pbmm ${destDeepSpeechFolder}/deepspeech.pbmm`, { shell: true })
-      await command(`mv -f ${tmpDir}/deepspeech-${deepSpeechVersion}-models.scorer ${destDeepSpeechFolder}/deepspeech.scorer`, { shell: true })
+      await command(`mv -f ${tmpDir}/model.tflite ${destCoquiFolder}/model.tflite`, { shell: true })
+      await command(`mv -f ${tmpDir}/huge-vocabulary.scorer ${destCoquiFolder}/huge-vocabulary.scorer`, { shell: true })
       log.success('Move done')
       log.success('Offline speech-to-text installed')
 

server/src/stt/coqui-stt/parser.js

@@ -0,0 +1,98 @@
+import wav from 'node-wav'
+import fs from 'fs'
+
+import log from '@/helpers/log'
+
+log.title('Coqui-ai Parser')
+
+const parser = { }
+let STT = { }
+
+/* istanbul ignore next */
+try {
+  STT = require('stt-gpu') // eslint-disable-line global-require, import/no-unresolved
+
+  log.success('GPU version found')
+} catch (eGpu) {
+  log.info('GPU version not found, trying to get the CPU version...')
+
+  try {
+    STT = require('stt') // eslint-disable-line global-require, import/no-unresolved
+
+    log.success('CPU version found')
+  } catch (eCpu) {
+    log.error(`No Coqui-ai library found:\nGPU: ${eGpu}\nCPU: ${eCpu}`)
+  }
+}
+
+let model = { }
+let desiredSampleRate = 16000
+
+/**
+ * Model and language model paths
+ */
+parser.conf = {
+  model: 'bin/coqui/model.tflite',
+  scorer: 'bin/coqui/huge-vocabulary.scorer'
+}
+
+/**
+ * Load models
+ */
+parser.init = (args) => {
+  log.info(`Loading model from file ${args.model}...`)
+
+  if (!fs.existsSync(args.model)) {
+    log.error(`Cannot find ${args.model}. You can setup the offline STT by running: "npm run setup:offline-stt"`)
+
+    return false
+  }
+
+  if (!fs.existsSync(args.scorer)) {
+    log.error(`Cannot find ${args.scorer}. You can setup the offline STT by running: "npm run setup:offline-stt"`)
+
+    return false
+  }
+
+  /* istanbul ignore if */
+  if (process.env.LEON_NODE_ENV !== 'testing') {
+    try {
+      model = new STT.Model(args.model)
+    } catch (error) { 
+      throw Error(`model.stt: ${error}`)
+    } 
+    desiredSampleRate = model.sampleRate()
+
+    try {
+      model.enableExternalScorer(args.scorer)
+    } catch (error) { 
+      throw Error(`model.enableExternalScorer: ${error}`)
+    } 
+  }
+
+  log.success('Model loaded')
+
+  return true
+}
+
+/**
+ * Parse file and infer
+ */
+parser.parse = (buffer, cb) => {
+  const wavDecode = wav.decode(buffer)
+
+  if (wavDecode.sampleRate < desiredSampleRate) {
+    log.warning(`Original sample rate (${wavDecode.sampleRate}) is lower than ${desiredSampleRate}Hz. Up-sampling might produce erratic speech recognition`)
+  }
+
+  /* istanbul ignore if */
+  if (process.env.LEON_NODE_ENV !== 'testing') {
+    const string = model.stt(buffer)
+
+    cb({ string })
+  }
+
+  return true
+}
+
+export default parser

server/src/stt/stt.js

@@ -10,7 +10,8 @@ class Stt {
     this.providers = [
       'deepspeech',
       'google-cloud-stt',
-      'watson-stt'
+      'watson-stt',
+      'coqui-stt'
     ]
     this.parser = { }