@absolutejs/voice-google-speech

@absolutejs/voice-google-speechv0.0.1-beta.6betaVoice & Media

Google Cloud Speech-to-Text adapter for @absolutejs/voice — buffered-batch (REST) + real-time streaming (gRPC-Web over HTTP/2, no @grpc/grpc-js dep)

#Installation

BASH

bun add @absolutejs/voice-google-speech

#Capabilities

Overview

Google Cloud Speech-to-Text adapter for @absolutejs/voice. Ships two complementary modes — pick whichever fits your latency / cost / infrastructure tradeoffs:

googleSpeech(...) — buffered-batch via REST /v1/speech:recognize. Accumulate PCM, POST on flush() / close(), get one final transcript per turn. Good for post-call analysis and short utterances; simplest auth (API key, OAuth, or refresh hook).

googleSpeechStream(...) — real-time bidirectional streaming via Google Speech v2 StreamingRecognize. Speaks gRPC over HTTP/2 directly (node:http2) with hand-rolled protobuf wire-format for the request/response messages and gRPC-Web framing. No @grpc/grpc-js dependency. Use this when you want partial transcripts as the caller is speaking.

Streaming (googleSpeechStream)

OAuth refresh (preferred for long-running deployments):

What this adapter does protocol-wise

Opens a single HTTP/2 stream against https://speech.googleapis.com/google.cloud.speech.v2.Speech/StreamingRecognize.

Show 10 more

Headers: Authorization: Bearer , Content-Type: application/grpc-web+proto, X-Grpc-Web: 1, TE: trailers.

Sends the first message as a StreamingRecognizeRequest carrying recognizer = projects/{project}/locations/{location}/recognizers/_ and a streaming_config derived from STTAdapterOpenOptions.format + your options.

Each call to session.send(audio) encodes a StreamingRecognizeRequest { audio = }, wraps it in a 5-byte gRPC frame, and writes it to the HTTP/2 stream.

Reads StreamingRecognizeResponse frames as they arrive, decodes results[].alternatives[0], and emits partial events while is_final = false, final events when is_final = true. result_end_offset is lifted to transcript.endedAtMs, language_code to transcript.language, confidence to transcript.confidence.

speech_event_type values of END_OF_SINGLE_UTTERANCE or SPEECH_ACTIVITY_END emit an endOfTurn (reason: "vendor").

Trailer frame (gRPC-Web high-bit flag) or HTTP/2 trailer with non-zero grpc-status emits an error event with code = and message "Google Speech gRPC status : ".

Streaming options

Option — Required — Default — Notes

accessToken / getAccessToken — one of — — — OAuth Bearer token (or async refresh hook). API keys are not supported on the streaming RPC; use the buffered-batch googleSpeech(...) for API-key flows.

project — yes — — — Google Cloud project id; combined with location to form the recognizer resource path.

Buffered-batch (googleSpeech)

See the existing buffered-batch section for the REST-based API key / OAuth flow. The streaming and batch exports can coexist in the same project; pick per-route.

Notes & limitations

Bun's fetch is half-duplex (request body must finish before the response body starts), which is why the streaming adapter uses node:http2 directly instead of fetch. Bun's node:http2 polyfill supports the bidirectional pattern this adapter needs.

The protobuf encoder/decoder is hand-rolled and covers only the fields actually used by StreamingRecognizeRequest/StreamingRecognizeResponse. If Google adds new fields you want surfaced (alternatives beyond [0], word-level timing, channel tags, etc.), open an issue.

For STTAdapterOpenOptions.format.encoding: pcm_s16le → LINEAR16, mulaw / pcm_mulaw → MULAW, alaw / pcm_alaw → ALAW. Other encodings are rejected at open time.

Show 1 more

session.close(reason) ends the request stream cleanly and emits close. Any in-flight transcripts that arrive after close are dropped.

Outcomes

What you can build

Build on the supported package contract

Use @absolutejs/voice-google-speech through its supported public entry points.

Hardening checklist

Production guidance

Make every external boundary explicitPin the deployed @absolutejs/voice-google-speech version, replace example or memory-backed dependencies with durable implementations, bound external calls, protect credentials, and emit enough evidence to retry or recover safely.

Follow in order

Troubleshooting path

Trace from the first failed boundary

Reproduce the smallest canonical @absolutejs/voice-google-speech example, confirm the supported entry point and version in the API explorer, then inspect the first boundary that did not produce its documented result.

#Streaming (googleSpeechStream)

Partial snippet

Working example for Streaming (googleSpeechStream).

import { voice } from "@absolutejs/voice";
import { googleSpeechStream } from "@absolutejs/voice-google-speech";

const app = voice({
  stt: googleSpeechStream({
    project: process.env.GOOGLE_PROJECT_ID!,
    accessToken: await mintBearerToken(),
    // optional:
    location: "global", // default
    model: "latest_long", // default; or 'latest_short' / 'chirp_2' / 'telephony' / ...
    language: "en-US",
    languages: ["en-US"], // used unless languageStrategy overrides
    interimResults: true, // default — partial transcripts emitted
    enableVoiceActivityEvents: true, // surface SPEECH_ACTIVITY_END as endOfTurn
    enableAutomaticPunctuation: true,
    enableWordConfidence: true,
  }),
});

#Streaming (googleSpeechStream) 2

Partial snippet

OAuth refresh (preferred for long-running deployments):

googleSpeechStream({
  project,
  getAccessToken: async () => await refreshGoogleAccessToken(),
});

#Public entry points

Supported entry points declared by this package manifest.

Package entry point declared in package.json.

@absolutejs/voice-google-speech@absolutejs/voice-google-speech/manifest@absolutejs/voice-google-speech/manifest.json

#Package commands

Scripts declared by this package manifest.

bun run buildrm -rf dist && bun build ./src/index.ts ./src/manifest.ts --outdir dist --target bun --external @absolutejs/manifest --external '@absolutejs/manifest/*' --external @absolutejs/voice --external '@absolutejs/voice/*' --external @sinclair/typebox --external '@sinclair/typebox/*' && tsc --emitDeclarationOnly --project tsconfig.json && absolute-manifest emit

bun run check:packagebun run format && bun run typecheck && bun run test && bun run verify-package && bun run build && bun run verify-package --artifacts

bun run formatprettier --write "./**/*.{js,ts,json,md}"

bun run testbun test

bun run typecheckbun run tsc --noEmit

#API reference

Search the declarations exported by the current package type files. Expand a symbol to inspect its source-backed signature.

11 symbols

googleSpeechexportPermalinkSource

googleSpeech

Exported from @absolutejs/voice-google-speech

Source README npm Voice Adapters overview

Previousvoice-gladia Nextvoice-lmnt