voice-router-dev 0.8.6 → 0.8.8

package/CHANGELOG.md

@@ -5,6 +5,71 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.8.7] - 2026-04-18
+
+### Added
+
+#### Speechmatics: Real-Time Streaming (`transcribeStream()`)
+
+Speechmatics now supports WebSocket-based real-time transcription via `wss://{region}.rt.speechmatics.com/v2`. The adapter follows the same pattern as Deepgram/Gladia/AssemblyAI streaming.
+
+**Protocol flow:**
+1. Connect with `Authorization: Bearer` header
+2. Send `StartRecognition` JSON with `audio_format` + `transcription_config`
+3. Wait for `RecognitionStarted` acknowledgment
+4. Stream binary audio frames via `sendAudio()`
+5. Receive `AddPartialTranscript` (partials) and `AddTranscript` (finals)
+6. `EndOfUtterance` boundaries trigger `onUtterance()` callback
+7. `EndOfStream` → `EndOfTranscript` for clean shutdown
+
+**Streaming options** (`speechmaticsStreaming`): `encoding`, `sampleRate`, `language`, `domain`, `operatingPoint`, `maxDelay`, `maxDelayMode`, `enablePartials`, `enableEntities`, `diarization`, `maxSpeakers`, `additionalVocab`, `conversationConfig`, `region`.
+
+**Type changes:**
+- `SpeechmaticsCapabilities.streaming` is now `true` — Speechmatics is included in `StreamingProviderType`
+- `SpeechmaticsStreamingOptions` added to `ProviderStreamingOptions` union and `StreamingOptionsForProvider<P>` conditional type
+- `StreamingOptions.speechmaticsStreaming` field added
+
+### Fixed
+
+#### Soniox: Fix Streaming WebSocket Initialization
+
+Three bugs in the Soniox streaming adapter:
+
+| Bug | Before (broken) | After (fixed) |
+|-----|-----------------|----------------|
+| **Init message** | Config sent as URL query params | JSON text frame sent after `ws.onopen` (Soniox requires first frame to be JSON) |
+| **Default model** | `stt-rt-preview` (deprecated/removed) | `stt-rt-v4` |
+| **Close detection** | 1s threshold for early-close detection | 5s threshold (Soniox takes ~3s to close) |
+
+The JSON init frame now includes `api_key`, `model`, `audio_format`, `sample_rate`, `num_channels`, and all optional config (diarization, language hints, context, etc.).
+
+#### Speechmatics: Fix Content-Type for URL-Based Batch Transcription
+
+Speechmatics `POST /v2/jobs` always requires `multipart/form-data`, but the URL path was sending a JSON body with `Content-Type: application/json`, causing HTTP 400 errors.
+
+The `config` field is now sent as a FormData field for both URL and file inputs. Also fixed the file upload path to properly convert `Buffer` to `Blob` before appending to FormData (pre-existing type error).
+
+#### Soniox: Migrate to Current Async Transcription API
+
+The batch transcription adapter was using the old `/speech/transcribe` endpoint which no longer exists (HTTP 404). Soniox migrated to an async job-based API.
+
+| | Before (broken) | After (fixed) |
+|---|---|---|
+| **Create job (URL)** | `POST /speech/transcribe` (JSON) | `POST /transcriptions` (JSON with `audio_url`) |
+| **Create job (file)** | `POST /speech/transcribe` (multipart) | `POST /files` → `POST /transcriptions` with `file_id` |
+| **Get result** | `GET /speech/transcripts/{id}` | `GET /transcriptions/{id}` (status) + `GET /transcriptions/{id}/transcript` (result) |
+| **Flow** | Synchronous (immediate result) | Async with `pollForCompletion()` |
+
+`normalizeResponse` updated to handle batch transcript tokens (no `is_final` field — all tokens are final) and read `audio_duration_ms` from job metadata.
+
+**No breaking changes for consumers.** The adapter's public API (`transcribe()`, `getTranscript()`) is unchanged.
+
+#### Azure STT: Add Utterance Extraction to Batch Transcription
+
+Azure batch transcription had words with speaker labels but wasn't building utterances from them. Now uses `buildUtterancesFromWords()` to group speaker-labeled words into utterances, matching all other adapters.
+
+---
+
 ## [0.8.6] - 2026-04-15
 
 ### Changed