Assistant answers general questions and stays silent for overheard talk ("Ignored." is not said); stt prompt setting sent as the transcription prompt

This commit is contained in:
Melvin Strobl
2026-11-06 08:04:19 +01:00
committed by stroblme
parent 7ddf8aa8f9
commit c80aad7dfe
6 changed files with 93 additions and 24 deletions
+23 -3
View File
@@ -186,7 +186,18 @@ class SttNode(_Speech):
alone.
"""
__slots__ = ("_heard", "_listening", "_lock", "_session", "_tasks")
class Params(BaseModel):
model_config = ConfigDict(extra="allow")
prompt: str = Field(
default="",
description=(
"Words the speech is likely to hold, such as the names of rooms "
"and devices, given to the server as its transcription prompt."
),
)
__slots__ = ("cfg", "_heard", "_listening", "_lock", "_session", "_tasks")
def __init__(
self,
@@ -195,6 +206,7 @@ class SttNode(_Speech):
params: dict[str, Any] | None = None,
name: str | None = None,
):
self.cfg = self.Params.model_validate(params or {})
self._lock = threading.Lock()
#: The chunks of the utterance being spoken, None between utterances.
self._session: asyncio.Queue[bytes | None] | None = None
@@ -293,7 +305,11 @@ class SttNode(_Speech):
try:
async with asyncio.timeout(UTTERANCE_S):
async for final, text in client.transcribe_stream(
url + "/v1", chunks, rate=RATE, idle_s=IDLE_S
url + "/v1",
chunks,
rate=RATE,
idle_s=IDLE_S,
prompt=self.cfg.prompt,
):
words = text.strip() if final else words + text
if final and words:
@@ -317,7 +333,11 @@ class SttNode(_Speech):
async def _clip(self, url: str, audio: dict[str, Any], data: bytes) -> None:
try:
text = await client.transcribe(
url + "/v1", data, "speech.wav", str(audio.get("media_type"))
url + "/v1",
data,
"speech.wav",
str(audio.get("media_type")),
prompt=self.cfg.prompt,
)
except Exception as exc:
logger.warning("'%s' lost a clip: %s: %s", self.id, type(exc).__name__, exc)