fix: 11Labs Scribe v2 model not working with EOT prediction model

livekit-plugins/livekit-plugins-elevenlabs/livekit/plugins/elevenlabs/languages.py

@@ -0,0 +1,114 @@
+"""
+Mapping of ISO 639-3 language codes used by Scribe to ISO 639-1
+used by Livekit turn detector multilingual model
+Note: ISO 639-3 language codes supported by Scribe but that have no ISO 639-1 code
+are defined with a None mapping value
+https://elevenlabs.io/docs/overview/capabilities/speech-to-text#supported-languages
+"""
+
+ISO_639_3_TO_1 = {
+    "afr": "af",
+    "amh": "am",
+    "ara": "ar",
+    "hye": "hy",
+    "asm": "as",
+    "ast": None,
+    "aze": "az",
+    "bel": "be",
+    "ben": "bn",
+    "bos": "bs",
+    "bul": "bg",
+    "mya": "my",
+    "yue": None,
+    "cat": "ca",
+    "ceb": None,
+    "nya": "ny",
+    "hrv": "hr",
+    "ces": "cs",
+    "dan": "da",
+    "nld": "nl",
+    "eng": "en",
+    "est": "et",
+    "fil": None,
+    "fin": "fi",
+    "fra": "fr",
+    "ful": "ff",
+    "glg": "gl",
+    "lug": "lg",
+    "kat": "ka",
+    "deu": "de",
+    "ell": "el",
+    "guj": "gu",
+    "hau": "ha",
+    "heb": "he",
+    "hin": "hi",
+    "hun": "hu",
+    "isl": "is",
+    "ibo": "ig",
+    "ind": "id",
+    "gle": "ga",
+    "ita": "it",
+    "jpn": "ja",
+    "jav": "jv",
+    "kea": None,
+    "kan": "kn",
+    "kaz": "kk",
+    "khm": "km",
+    "kor": "ko",
+    "kur": "ku",
+    "kir": "ky",
+    "lao": "lo",
+    "lav": "lv",
+    "lin": "ln",
+    "lit": "lt",
+    "luo": None,
+    "ltz": "lb",
+    "mkd": "mk",
+    "msa": "ms",
+    "mal": "ml",
+    "mlt": "mt",
+    "zho": "zh",
+    "mri": "mi",
+    "mar": "mr",
+    "mon": "mn",
+    "nep": "ne",
+    "nso": None,
+    "nor": "no",
+    "oci": "oc",
+    "ori": "or",
+    "pus": "ps",
+    "fas": "fa",
+    "pol": "pl",
+    "por": "pt",
+    "pan": "pa",
+    "ron": "ro",
+    "rus": "ru",
+    "srp": "sr",
+    "sna": "sn",
+    "snd": "sd",
+    "slk": "sk",
+    "slv": "sl",
+    "som": "so",
+    "spa": "es",
+    "swa": "sw",
+    "swe": "sv",
+    "tam": "ta",
+    "tgk": "tg",
+    "tel": "te",
+    "tha": "th",
+    "tur": "tr",
+    "ukr": "uk",
+    "umb": None,
+    "urd": "ur",
+    "uzb": "uz",
+    "vie": "vi",
+    "cym": "cy",
+    "wol": "wo",
+    "xho": "xh",
+    "zul": "zu",
+}
+
+
+def iso639_3_to_1(code: str | None) -> str | None:
+    """Convert ISO 639-3 to ISO 639-1. Returns None if no mapping exists."""
+    return ISO_639_3_TO_1.get(code.lower()) if code else None

livekit-plugins/livekit-plugins-elevenlabs/livekit/plugins/elevenlabs/stt.py

@@ -39,6 +39,7 @@
 from livekit.agents.utils import AudioBuffer, http_context, is_given
 from livekit.agents.voice.io import TimedString
 
+from .languages import iso639_3_to_1
 from .log import logger
 from .models import STTRealtimeSampleRates
 
@@ -220,8 +221,9 @@ async def _recognize_impl(
         except Exception as e:
             raise APIConnectionError() from e
 
+        normalized_language = iso639_3_to_1(language_code) or language_code
         return self._transcription_to_speech_event(
-            language_code=language_code,
+            language_code=normalized_language,
             text=extracted_text,
             start_time=start_time,
             end_time=end_time,
@@ -483,9 +485,10 @@ def _process_stream_event(self, data: dict) -> None:
         start_time = words[0].get("start", 0) if words else 0
         end_time = words[-1].get("end", 0) if words else 0
 
+        normalized_language = iso639_3_to_1(self._language) or self._language or "en"
         # 11labs only sends word timestamps for final transcripts
         speech_data = stt.SpeechData(
-            language=self._language or "en",
+            language=normalized_language,
             text=text,
             start_time=start_time + self.start_time_offset,
             end_time=end_time + self.start_time_offset,