ElevenLabs STT stream ignores sample rate configuration

[ameyakhare]

Bug Description

The elevenlabs STT implementation passes the sample rate/audio format in the wrong parameter during wss connection. The correct parameter is audio_format, but the format is being passed in the encoding (non-existent) parameter. API Reference here.

Expected Behavior

The sample rate/encoding should be passed in the audio_format parameter during wss connection.

Reproduction Steps

elevenlabs.STT(model="scribe_v2_realtime", sample_rate=8000)

pcm_8000 is passed to the non-existent encoding parameter. This defaults to pcm_16000, resulting in incorrect transcription timestamps, silence length detection, min silence/speech threshold detection, etc. for vad-based commits

Operating System

macOS Tahoe

Models Used

ElevenLabs Scribe V2 Realtime

Package Versions

livekit-plugins-elevenlabs==1.5.6

Session/Room/Call IDs

No response

Proposed Solution

--- a/livekit-plugins/livekit-plugins-elevenlabs/livekit/plugins/elevenlabs/stt.py
+++ b/livekit-plugins/livekit-plugins-elevenlabs/livekit/plugins/elevenlabs/stt.py
@@ -365,7 +365,6 @@ class SpeechStream(stt.SpeechStream):
                                 "message_type": "input_audio_chunk",
                                 "audio_base_64": "",
                                 "commit": False,
-                                "sample_rate": self._opts.sample_rate,
                             }
                         )
                     )
@@ -403,7 +402,6 @@ class SpeechStream(stt.SpeechStream):
                                 "message_type": "input_audio_chunk",
                                 "audio_base_64": audio_b64,
                                 "commit": False,
-                                "sample_rate": self._opts.sample_rate,
                             }
                         )
                     )
@@ -484,7 +482,7 @@ class SpeechStream(stt.SpeechStream):
         commit_strategy = "manual" if self._opts.server_vad is None else "vad"
         params = [
             f"model_id={self._opts.model_id}",
-            f"encoding=pcm_{self._opts.sample_rate}",
+            f"audio_format=pcm_{self._opts.sample_rate}",
             f"commit_strategy={commit_strategy}",
         ]

Additional Context

No response

Screenshots and Recordings

No response