utils

Utility functions for extracting probability metrics from STT services.

pipecat.services.whisper.utils.extract_whisper_probability(frame: TranscriptionFrame) → float | None[source]

Extract probability from Whisper-based TranscriptionFrame result.

Works with Groq, OpenAI Whisper, or other Whisper-based services that use verbose_json format with segments containing avg_logprob.

Converts avg_logprob to probability.

Parameters:: frame – TranscriptionFrame with result from GroqSTTService or OpenAISTTService (when include_prob_metrics=True and using Whisper models).
Returns:: Probability (0-1) if available, None otherwise.

Example:

from pipecat.services.groq.stt import GroqSTTService
from pipecat.services.whisper.utils import extract_whisper_probability

stt = GroqSTTService(include_prob_metrics=True)
# ... use stt in pipeline ...
# In your frame processor:
if isinstance(frame, TranscriptionFrame):
    prob = extract_whisper_probability(frame)
    if prob:
        print(f"Transcription confidence: {prob:.2%}")

pipecat.services.whisper.utils.extract_openai_gpt4o_probability(frame: TranscriptionFrame) → float | None[source]

Extract probability from OpenAI GPT-4o-transcribe TranscriptionFrame result.

Parameters:: frame – TranscriptionFrame with result from OpenAISTTService using GPT-4o-transcribe model (when include_prob_metrics=True).
Returns:: Probability (0-1) if available, None otherwise.

Example:

from pipecat.services.openai.stt import OpenAISTTService
from pipecat.services.whisper.utils import extract_openai_gpt4o_probability

stt = OpenAISTTService(model="gpt-4o-transcribe", include_prob_metrics=True)
# ... use stt in pipeline ...
# In your frame processor:
if isinstance(frame, TranscriptionFrame):
    prob = extract_openai_gpt4o_probability(frame)
    if prob:
        print(f"Transcription confidence: {prob:.2%}")

pipecat.services.whisper.utils.extract_deepgram_probability(frame: TranscriptionFrame) → float | None[source]

Extract probability from Deepgram TranscriptionFrame result.

Parameters:: frame – TranscriptionFrame with result from DeepgramSTTService.
Returns:: Probability (0-1) if available, None otherwise. Returns alternative-level confidence if available, otherwise calculates average confidence from word-level confidences.

Example:

from pipecat.services.deepgram.stt import DeepgramSTTService
from pipecat.services.whisper.utils import extract_deepgram_probability

stt = DeepgramSTTService()
# ... use stt in pipeline ...
# In your frame processor:
if isinstance(frame, TranscriptionFrame):
    prob = extract_deepgram_probability(frame)
    if prob:
        print(f"Transcription confidence: {prob:.2%}")