feat: Add ISO-639-1 language code support for all backends

BACKENDS.md

@@ -84,7 +84,6 @@ Uses Deepgram's WebSocket streaming API for real-time transcription.
 DEEPGRAM_API_KEY=your-key-here
 DEEPGRAM_MODEL=nova-2
 DEEPGRAM_ENCODING=linear16        # Audio encoding: linear16 (PCM) or opus (default: linear16)
-DEEPGRAM_LANGUAGE=multi           # Multilingual code-switching (default)
 DEEPGRAM_INCLUDE_LANGUAGE=true    # Append language to transcript (e.g., "Hello [en]")
 DEEPGRAM_PUNCTUATE=true
 DEEPGRAM_DIARIZE=false
@@ -109,8 +108,8 @@ PROVIDERS_PRIORITY=deepgram,openai,gemini
 - Supports KeepAlive, Finalize, and CloseStream control messages
 - Authentication via Sec-WebSocket-Protocol header
 - Multilingual streaming support:
-  - Defaults to `language=multi` for automatic multilingual code-switching (31+ languages with Nova-3)
-  - Can specify single language (e.g., `en`, `es`, `fr`, `de`, `pt`, etc.)
+  - Defaults to `language=multi` (auto-detect) when no `lang` URL parameter is provided
+  - Specify language via URL: `?lang=en`, `?lang=es`, `?lang=fr`, etc. (ISO-639-1 codes)
   - Automatically adds `endpointing=100` for multilingual mode (recommended for code-switching)
   - Optional: Include detected language in transcript (e.g., `"Hello [en]"`) via `DEEPGRAM_INCLUDE_LANGUAGE=true`
   - **Note**: `detect_language` parameter is NOT supported for streaming (only for pre-recorded audio)

README.md

@@ -82,7 +82,6 @@ Set environment variables or use a `.env` file:
 |----------|---------|-------------|
 | `OPENAI_MODEL` | `gpt-4o-mini-transcribe` | OpenAI model |
 | `DEEPGRAM_MODEL` | `nova-2` | Deepgram model |
-| `DEEPGRAM_LANGUAGE` | `multi` | Language code or `multi` for auto |
 | `DEEPGRAM_ENCODING` | `linear16` | `linear16`, `opus`, or `ogg-opus` |
 | `GEMINI_MODEL` | `gemini-2.0-flash-exp` | Gemini model |
 
@@ -133,7 +132,7 @@ ws://host:port/transcribe?sessionId=xxx&sendBack=true
 | `sendBackInterim` | `false` | Return interim transcriptions |
 | `provider` | (auto) | Override provider selection |
 | `encoding` | `opus` | Audio encoding: `opus` or `ogg-opus` |
-| `lang` | (auto) | Language hint |
+| `lang` | (auto) | ISO-639-1 language code (e.g., `en`, `de`, `fr`). Auto-detect if omitted. |
 
 ### Client Messages
 

src/backends/DeepgramBackend.ts

@@ -70,15 +70,13 @@ export class DeepgramBackend implements TranscriptionBackend {
 				}
 
 				// Language configuration
-				// Use per-connection language if specified, otherwise use global config
-				const language = backendConfig.language || config.deepgram.language;
-				if (language) {
-					params.set('language', language);
-					// For multilingual streaming, add recommended endpointing
-					// See: https://developers.deepgram.com/docs/multilingual-code-switching
-					if (language === 'multi') {
-						params.set('endpointing', '100');
-					}
+				// Use per-connection language if specified, otherwise auto-detect (multi)
+				const language = backendConfig.language || 'multi';
+				params.set('language', language);
+				// For multilingual streaming, add recommended endpointing
+				// See: https://developers.deepgram.com/docs/multilingual-code-switching
+				if (language === 'multi') {
+					params.set('endpointing', '100');
 				}
 
 				// Note: detect_language is NOT supported for streaming

src/backends/GeminiBackend.ts

@@ -10,6 +10,7 @@ import logger from '../logger';
 import type { TranscriptionBackend, BackendConfig } from './TranscriptionBackend';
 import type { TranscriptionMessage } from '../transcriberproxy';
 import { writeMetric } from '../metrics';
+import { toLanguageName } from '../languageMap';
 
 // Gemini WebSocket API endpoint (v1beta - more stable than v1alpha)
 const GEMINI_WS_BASE = 'wss://generativelanguage.googleapis.com/ws/google.ai.generativelanguage.v1beta.GenerativeService.BidiGenerateContent';
@@ -163,7 +164,8 @@ export class GeminiBackend implements TranscriptionBackend {
 		let systemInstruction = this.backendConfig.prompt || 'Transcribe the following audio. Output only the transcribed text.';
 
 		if (this.backendConfig.language) {
-			systemInstruction += ` The audio is in ${this.backendConfig.language}.`;
+			const languageName = toLanguageName(this.backendConfig.language);
+			systemInstruction += ` The audio is in ${languageName}.`;
 		}
 
 		const setupMessage = {

src/config.ts

@@ -52,7 +52,6 @@ export const config = {
 	deepgram: {
 		apiKey: process.env.DEEPGRAM_API_KEY || '',
 		model: process.env.DEEPGRAM_MODEL || 'nova-2',
-		language: process.env.DEEPGRAM_LANGUAGE || 'multi',
 		encoding: (process.env.DEEPGRAM_ENCODING || 'linear16') as 'opus' | 'linear16',
 		punctuate: process.env.DEEPGRAM_PUNCTUATE === 'true',
 		diarize: process.env.DEEPGRAM_DIARIZE === 'true',

src/languageMap.ts

@@ -0,0 +1,68 @@
+/**
+ * Language code mapping for transcription backends
+ *
+ * Accepts ISO-639-1 short codes (e.g., "en", "de", "fr") and converts
+ * them to the format required by each backend.
+ */
+
+// ISO-639-1 to full language name mapping (for Gemini prompts)
+const ISO_TO_NAME: Record<string, string> = {
+	af: 'Afrikaans',
+	ar: 'Arabic',
+	be: 'Belarusian',
+	bg: 'Bulgarian',
+	bn: 'Bengali',
+	bs: 'Bosnian',
+	ca: 'Catalan',
+	cs: 'Czech',
+	da: 'Danish',
+	de: 'German',
+	el: 'Greek',
+	en: 'English',
+	es: 'Spanish',
+	et: 'Estonian',
+	fa: 'Persian',
+	fi: 'Finnish',
+	fr: 'French',
+	he: 'Hebrew',
+	hi: 'Hindi',
+	hr: 'Croatian',
+	hu: 'Hungarian',
+	id: 'Indonesian',
+	it: 'Italian',
+	ja: 'Japanese',
+	kn: 'Kannada',
+	ko: 'Korean',
+	lt: 'Lithuanian',
+	lv: 'Latvian',
+	mk: 'Macedonian',
+	mr: 'Marathi',
+	ms: 'Malay',
+	nl: 'Dutch',
+	no: 'Norwegian',
+	pl: 'Polish',
+	pt: 'Portuguese',
+	ro: 'Romanian',
+	ru: 'Russian',
+	sk: 'Slovak',
+	sl: 'Slovenian',
+	sr: 'Serbian',
+	sv: 'Swedish',
+	ta: 'Tamil',
+	te: 'Telugu',
+	th: 'Thai',
+	tl: 'Tagalog',
+	tr: 'Turkish',
+	uk: 'Ukrainian',
+	ur: 'Urdu',
+	vi: 'Vietnamese',
+	zh: 'Chinese',
+};
+
+/**
+ * Convert ISO-639-1 code to full language name (for Gemini).
+ * Returns the code itself if not found in mapping.
+ */
+export function toLanguageName(code: string): string {
+	return ISO_TO_NAME[code.toLowerCase()] || code;
+}