voice-router-dev 0.8.3 → 0.8.5

package/CHANGELOG.md

@@ -5,10 +5,116 @@ 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).
 
-## [0.9.1] - 2026-04-05
+## [0.8.5] - 2026-04-07
+
+### Changed
+
+#### AssemblyAI: Migrate to Live Docs Spec (100% OpenAPI Pipeline)
+
+AssemblyAI deleted their `assemblyai-api-spec` GitHub repo, breaking `pnpm openapi:sync` with 404 errors. The spec source is now the live Fern-generated docs spec at `https://www.assemblyai.com/docs/openapi.json`.
+
+**New pipeline** (follows the same pattern as ElevenLabs/OpenAI):
+
+```
+assemblyai.com/docs/openapi.json → fix-assemblyai-spec.js → Orval → adapter
+```
+
+**`fix-assemblyai-spec.js`** normalizes the docs spec before Orval generation:
+1. Filters paths to `/v2/*` STT endpoints (drops `/chat/completions`, `/understanding`, `/v3/token`)
+2. Maps operationIds back to descriptive names (`submit` → `createTranscript`, `get` → `getTranscript`, etc.)
+3. Strips inline `Authorization` header parameters (Orval handles auth via security schemes)
+4. Collects `$ref` references, resolves transitive deps, removes unreferenced schemas
+5. Fixes malformed array schemas and missing path parameters
+6. Strips deprecated `speech_model` from `TranscriptParams` (see fix below)
+
+**Schema changes from new spec:**
+
+| Aspect | Old (v1.3.4, dead repo) | New (docs, live) |
+|--------|------------------------|-------------------|
+| `TranscriptParams` | Split: `TranscriptParams` + `TranscriptOptionalParams` | Single merged `TranscriptParams` |
+| `language_code` | `string` | `TranscriptParamsLanguageCode` (proper enum) |
+| `ListTranscriptParams` | Separate schema + 5 scalar types | Inline query params on `GET /v2/transcript` |
+| LeMUR schemas | 7 schemas | Dropped (not STT) |
+| Webhook notification schemas | 3 schemas | Dropped (handler uses local types) |
+
+**Deleted:** `scripts/fix-assemblyai-missing-schemas.js` — the 5 scalar types (`AfterId`, `BeforeId`, `CreatedOn`, `Limit`, `ThrottledOnly`) are now inline query parameters in the new spec.
+
+**No breaking changes for consumers.** API function names (`createTranscript`, `getTranscript`, etc.) are preserved via operationId mapping. The `assemblyai` passthrough options type changed from `Partial<TranscriptOptionalParams>` to `Partial<TranscriptParams>` — all fields from the old type exist in the new one.
+
+### Fixed
+
+#### AssemblyAI: Strip Deprecated `speech_model` from Request Schema
+
+The docs spec includes both `speech_model` (singular, deprecated) and `speech_models` (plural, required array) on `TranscriptParams`. The API rejects requests containing both fields with HTTP 400:
+
+```
+"speech_model and speech_models cannot be used in the same request"
+```
+
+**Two-level fix:**
+
+1. **Spec level:** `fix-assemblyai-spec.js` strips `speech_model` from `TranscriptParams` before Orval generates types. This prevents it from appearing in generated Zod schemas and DynamicFieldForm UI.
+
+2. **Adapter level:** If users pass `speech_model` via passthrough options (e.g., from older code), the adapter gracefully migrates it to `speech_models: [value]` instead of sending both fields.
+
+```typescript
+// This still works — adapter migrates automatically
+{ assemblyai: { speech_model: "universal-3-pro" } }
+// → sent as: { speech_models: ["universal-3-pro"] }
+
+// Preferred: use the new field directly
+{ assemblyai: { speech_models: ["universal-3-pro", "universal-2"] } }
+```
+
+**Note:** `speech_model` is preserved in the `Transcript` response schema (read-only — tells you which model was used).
+
+---
+
+## [0.8.4] - 2026-04-06
 
 ### Fixed
 
+#### AssemblyAI: Migrate `speech_model` → `speech_models` (API Breaking Change)
+
+AssemblyAI deprecated the singular `speech_model` parameter and now requires `speech_models` (plural, array). The old field is rejected with HTTP 400:
+
+```
+"speech_models" must be a non-empty list containing one or more of: "universal-3-pro", "universal-2"
+```
+
+**What changed:**
+
+| | Before (0.8.3) | After (0.8.4) |
+|---|---|---|
+| **API field** | `speech_model: "best"` (singular, deprecated) | `speech_models: ["universal-3-pro"]` (array, required) |
+| **Model values** | `best`, `slam-1`, `universal` | `universal-3-pro`, `universal-2` |
+| **Constants** | `AssemblyAITranscriptionModel.best` | `AssemblyAITranscriptionModel["universal-3-pro"]` |
+
+**Adapter fix:** `options.model` now maps to `request.speech_models = [model]` instead of `request.speech_model = model`.
+
+**Generated types regenerated** from AssemblyAI's updated OpenAPI spec (v1.3.4):
+- `SpeechModel` is now `string` (no enum — AssemblyAI removed the fixed list)
+- `TranscriptOptionalParams.speech_models` added (required `SpeechModel[]`)
+- `TranscriptOptionalParams.speech_model` marked `@deprecated`
+- New response field: `speech_model_used` (which model actually ran)
+
+**Migration:**
+
+```typescript
+// Before
+import { AssemblyAITranscriptionModel } from 'voice-router-dev/constants'
+{ model: AssemblyAITranscriptionModel.best }
+
+// After
+import { AssemblyAITranscriptionModel } from 'voice-router-dev/constants'
+{ model: AssemblyAITranscriptionModel["universal-3-pro"] }
+
+// Or pass directly via assemblyai-specific options for multi-model routing
+{ assemblyai: { speech_models: ["universal-3-pro", "universal-2"] } }
+```
+
+**Note:** Streaming is unaffected — it still uses `speech_model` query parameter with streaming-specific model names (`universal-streaming-english`, `universal-streaming-multilingual`).
+
 #### Unified Error Normalization Across All Providers
 
 HTTP errors from all 8 providers now return semantic error codes and the actual provider error message instead of axios internals.

package/dist/constants.d.mts

@@ -2792,25 +2792,24 @@ declare const AssemblyAIEncoding: {
 /**
  * AssemblyAI batch transcription models
  *
- * Values: `best`, `slam-1`, `universal`
+ * Uses the `speech_models` (plural) API parameter — pass as array.
+ * AssemblyAI routes audio to the best available model from the list.
  *
- * - `best`: Highest accuracy, best for most use cases (default)
- * - `slam-1`: Speech-Language Aligned Model, optimized for specific domains
- * - `universal`: General-purpose model with broad language support
+ * - `universal-3-pro`: Highest accuracy, latest generation
+ * - `universal-2`: Previous generation, broad language support
  *
  * @example
  * ```typescript
  * import { AssemblyAITranscriptionModel } from 'voice-router-dev/constants'
  *
  * await router.transcribe('assemblyai', audioUrl, {
- *   speechModel: AssemblyAITranscriptionModel.best
+ *   model: AssemblyAITranscriptionModel["universal-3-pro"]
  * })
  * ```
  */
 declare const AssemblyAITranscriptionModel: {
-    readonly best: "best";
-    readonly "slam-1": "slam-1";
-    readonly universal: "universal";
+    readonly "universal-3-pro": "universal-3-pro";
+    readonly "universal-2": "universal-2";
 };
 /**
  * AssemblyAI language codes for transcription

package/dist/constants.d.ts

@@ -2792,25 +2792,24 @@ declare const AssemblyAIEncoding: {
 /**
  * AssemblyAI batch transcription models
  *
- * Values: `best`, `slam-1`, `universal`
+ * Uses the `speech_models` (plural) API parameter — pass as array.
+ * AssemblyAI routes audio to the best available model from the list.
  *
- * - `best`: Highest accuracy, best for most use cases (default)
- * - `slam-1`: Speech-Language Aligned Model, optimized for specific domains
- * - `universal`: General-purpose model with broad language support
+ * - `universal-3-pro`: Highest accuracy, latest generation
+ * - `universal-2`: Previous generation, broad language support
  *
  * @example
  * ```typescript
  * import { AssemblyAITranscriptionModel } from 'voice-router-dev/constants'
  *
  * await router.transcribe('assemblyai', audioUrl, {
- *   speechModel: AssemblyAITranscriptionModel.best
+ *   model: AssemblyAITranscriptionModel["universal-3-pro"]
  * })
  * ```
  */
 declare const AssemblyAITranscriptionModel: {
-    readonly best: "best";
-    readonly "slam-1": "slam-1";
-    readonly universal: "universal";
+    readonly "universal-3-pro": "universal-3-pro";
+    readonly "universal-2": "universal-2";
 };
 /**
  * AssemblyAI language codes for transcription

package/dist/constants.js

@@ -2982,13 +2982,6 @@ var TranslationLanguageCodeEnum = {
   zh: "zh"
 };
 
-// src/generated/assemblyai/schema/speechModel.ts
-var SpeechModel = {
-  best: "best",
-  "slam-1": "slam-1",
-  universal: "universal"
-};
-
 // src/generated/assemblyai/schema/transcriptLanguageCode.ts
 var TranscriptLanguageCode = {
   en: "en",
@@ -3372,7 +3365,10 @@ var AssemblyAIEncoding = {
   /** μ-law (telephony) */
   pcmMulaw: "pcm_mulaw"
 };
-var AssemblyAITranscriptionModel = SpeechModel;
+var AssemblyAITranscriptionModel = {
+  "universal-3-pro": "universal-3-pro",
+  "universal-2": "universal-2"
+};
 var AssemblyAILanguage = TranscriptLanguageCode;
 var AssemblyAISpeechModel = {
   /** Optimized for English */

package/dist/constants.mjs

@@ -2877,13 +2877,6 @@ var TranslationLanguageCodeEnum = {
   zh: "zh"
 };
 
-// src/generated/assemblyai/schema/speechModel.ts
-var SpeechModel = {
-  best: "best",
-  "slam-1": "slam-1",
-  universal: "universal"
-};
-
 // src/generated/assemblyai/schema/transcriptLanguageCode.ts
 var TranscriptLanguageCode = {
   en: "en",
@@ -3267,7 +3260,10 @@ var AssemblyAIEncoding = {
   /** μ-law (telephony) */
   pcmMulaw: "pcm_mulaw"
 };
-var AssemblyAITranscriptionModel = SpeechModel;
+var AssemblyAITranscriptionModel = {
+  "universal-3-pro": "universal-3-pro",
+  "universal-2": "universal-2"
+};
 var AssemblyAILanguage = TranscriptLanguageCode;
 var AssemblyAISpeechModel = {
   /** Optimized for English */