Speech vendor/cobalt (#463)

* initial changes for cobalt speech

* wip

* wip

* update to drachtio-fsmrf that supports cobalt

* update to verb-specifications with cobalt speech support

* more wip

* lint

* use node 18 with gh actions

* support for compiling cobalt hints

* fix bug in uuid_cobalt_compile_context

* update verb-specifications

* remove repeated code

* cobalt support for transcribe

* update to verb specs
This commit is contained in:
Dave Horton
2023-09-13 09:47:30 -04:00
committed by GitHub
parent d220733dea
commit a1793ac359
11 changed files with 204 additions and 24 deletions

View File

@@ -9,6 +9,7 @@ const {
NvidiaTranscriptionEvents,
IbmTranscriptionEvents,
SonioxTranscriptionEvents,
CobaltTranscriptionEvents,
DeepgramTranscriptionEvents,
JambonzTranscriptionEvents,
AmdEvents,
@@ -314,6 +315,10 @@ module.exports = (logger) => {
ep.addCustomEventListener(NvidiaTranscriptionEvents.Transcription, amd.transcriptionHandler);
break;
case 'cobalt':
ep.addCustomEventListener(CobaltTranscriptionEvents.Transcription, amd.transcriptionHandler);
break;
default:
if (vendor.startsWith('custom:')) {
ep.addCustomEventListener(JambonzTranscriptionEvents.Transcription, amd.transcriptionHandler);

View File

@@ -95,6 +95,11 @@
"Transcription": "soniox_transcribe::transcription",
"Error": "soniox_transcribe::error"
},
"CobaltTranscriptionEvents": {
"Transcription": "cobalt_speech::transcription",
"CompileContext": "cobalt_speech::compile_context_response",
"Error": "cobalt_speech::error"
},
"IbmTranscriptionEvents": {
"Transcription": "ibm_transcribe::transcription",
"ConnectFailure": "ibm_transcribe::connect_failed",

View File

@@ -81,6 +81,10 @@ const speechMapper = (cred) => {
const o = JSON.parse(decrypt(credential));
obj.riva_server_uri = o.riva_server_uri;
}
else if ('cobalt' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.cobalt_server_uri = o.cobalt_server_uri;
}
else if (obj.vendor.startsWith('custom:')) {
const o = JSON.parse(decrypt(credential));
obj.auth_token = o.auth_token;

View File

@@ -7,6 +7,7 @@ const {
DeepgramTranscriptionEvents,
SonioxTranscriptionEvents,
NvidiaTranscriptionEvents,
CobaltTranscriptionEvents,
JambonzTranscriptionEvents
} = require('./constants');
@@ -92,6 +93,11 @@ const stickyVars = {
nvidia: [
'NVIDIA_HINTS'
],
cobalt: [
'COBALT_SPEECH_HINTS',
'COBALT_COMPILED_CONTEXT_DATA',
'COBALT_METADATA'
],
soniox: [
'SONIOX_PROFANITY_FILTER',
'SONIOX_MODEL'
@@ -225,6 +231,26 @@ const normalizeGoogle = (evt, channel, language) => {
};
};
const normalizeCobalt = (evt, channel, language) => {
const copy = JSON.parse(JSON.stringify(evt));
const alternatives = (evt.alternatives || [])
.map((alt) => ({
confidence: alt.confidence,
transcript: alt.transcript_formatted,
}));
return {
language_code: language,
channel_tag: channel,
is_final: evt.is_final,
alternatives,
vendor: {
name: 'cobalt',
evt: copy
}
};
};
const normalizeCustom = (evt, channel, language, vendor) => {
const copy = JSON.parse(JSON.stringify(evt));
return {
@@ -317,6 +343,8 @@ module.exports = (logger) => {
return normalizeNvidia(evt, channel, language);
case 'soniox':
return normalizeSoniox(evt, channel, language);
case 'cobalt':
return normalizeCobalt(evt, channel, language);
default:
if (vendor.startsWith('custom:')) {
return normalizeCustom(evt, channel, language, vendor);
@@ -584,6 +612,25 @@ module.exports = (logger) => {
{NVIDIA_CUSTOM_CONFIGURATION: JSON.stringify(nvidiaOptions.customConfiguration)}),
};
}
else if ('cobalt' === vendor) {
const {cobaltOptions = {}} = rOpts;
const cobaltUri = cobaltOptions.serverUri || sttCredentials.cobalt_server_uri;
opts = {
...opts,
...(rOpts.words && {COBALT_WORD_TIME_OFFSETS: 1}),
...(!rOpts.words && {COBALT_WORD_TIME_OFFSETS: 0}),
...(rOpts.model && {COBALT_MODEL: rOpts.model}),
...(cobaltUri && {COBALT_SERVER_URI: cobaltUri}),
...(rOpts.hints?.length > 0 && typeof rOpts.hints[0] === 'string' &&
{COBALT_SPEECH_HINTS: rOpts.hints.join(',')}),
...(rOpts.hints?.length > 0 && typeof rOpts.hints[0] === 'object' &&
{COBALT_SPEECH_HINTS: JSON.stringify(rOpts.hints)}),
...(rOpts.hints?.length > 0 && {COBALT_CONTEXT_TOKEN: cobaltOptions.contextToken || 'unk:default'}),
...(cobaltOptions.metadata && {COBALT_METADATA: cobaltOptions.metadata}),
...(cobaltOptions.enableConfusionNetwork && {COBALT_ENABLE_CONFUSION_NETWORK: 1}),
...(cobaltOptions.compiledContextData && {COBALT_COMPILED_CONTEXT_DATA: cobaltOptions.compiledContextData}),
};
}
else if (vendor.startsWith('custom:')) {
let {options = {}} = rOpts;
const {auth_token, custom_stt_url} = sttCredentials;
@@ -633,6 +680,9 @@ module.exports = (logger) => {
ep.removeCustomEventListener(SonioxTranscriptionEvents.Transcription);
ep.removeCustomEventListener(CobaltTranscriptionEvents.Transcription);
ep.removeCustomEventListener(CobaltTranscriptionEvents.CompileContext);
ep.removeCustomEventListener(NvidiaTranscriptionEvents.Transcription);
ep.removeCustomEventListener(NvidiaTranscriptionEvents.TranscriptionComplete);
ep.removeCustomEventListener(NvidiaTranscriptionEvents.StartOfSpeech);
@@ -664,6 +714,10 @@ module.exports = (logger) => {
const {apiKey} = recognizer.sonioxOptions || {};
if (apiKey) return {api_key: apiKey};
}
else if (recognizer.vendor === 'cobalt') {
const {serverUri} = recognizer.cobaltOptions || {};
if (serverUri) return {cobalt_server_uri: serverUri};
}
else if (recognizer.vendor === 'ibm') {
const {ttsApiKey, ttsRegion, sttApiKey, sttRegion, instanceId} = recognizer.ibmOptions || {};
if (ttsApiKey || sttApiKey) return {