pipecat-plugins-pinch 0.1.0

Pinch real-time speech-to-speech translation plugin for Pipecat

pipecat-plugins-pinch

Real-time voice translation for Pipecat powered by Pinch.

---

Installation

pip install pipecat-plugins-pinch

Requires Python ≥ 3.10.

---

Prerequisites

You need a Pinch API key. Get one at the developers portal.

Set it in your environment:

export PINCH_API_KEY=pk_your_key_here

---

How it fits into a Pipecat pipeline

PinchTranslatorService is a drop-in FrameProcessor. It sits between your transport's input and output — receiving InputAudioRawFrame from the user and emitting OutputAudioRawFrame (translated speech) and TranscriptionFrame (transcripts) downstream.

transport.input()
      │
      ▼
PinchTranslatorService        ← translate en-US → es-ES
      │
      ├─► OutputAudioRawFrame  → transport.output()  (translated audio)
      └─► TranscriptionFrame   → your handler        (transcripts)

---

Usage

from pipecat.pipeline.pipeline import Pipeline
from pipecat.pipeline.runner import PipelineRunner
from pipecat.pipeline.task import PipelineParams, PipelineTask
from pipecat.plugins.pinch import PinchTranslatorService, TranslatorOptions

async def main():
    # Replace with your preferred Pipecat transport (Daily, LiveKit, WebSocket, etc.)
    # transport = DailyTransport(...)

    translator = PinchTranslatorService(
        options=TranslatorOptions(
            source_language="en-US",
            target_language="es-ES",
            voice_type="clone",   # preserves the speaker's voice
        ),
        # api_key="pk_..."  ← or set PINCH_API_KEY env var
    )

    pipeline = Pipeline([
        transport.input(),
        translator,
        transport.output(),
    ])

    task = PipelineTask(pipeline, params=PipelineParams(allow_interruptions=True))
    await PipelineRunner().run(task)

The plugin handles everything internally — calling the Pinch API, managing the translation session, resampling audio, and routing frames — so you don't need to configure anything beyond TranslatorOptions.

---

Configuration

TranslatorOptions

Parameter	Type	Default	Description
source_language	str	required	code for the speaker's language (e.g. "en-US")
target_language	str	required	code for the output language (e.g. "es-ES")
voice_type	str	"clone"	Voice used for translated output: "clone", "female", or "male"

PinchTranslatorService

Parameter	Type	Default	Description
options	TranslatorOptions	required	Language and voice configuration
api_key	str | None	None	Pinch API key. Falls back to PINCH_API_KEY env var

---

Supported languages

Full list of language codes: supported languages

---

Transcript frames

The plugin emits standard Pipecat transcript frames for both the original and translated speech. You can consume these in any downstream FrameProcessor.

from pipecat.frames.frames import TranscriptionFrame, InterimTranscriptionFrame
from pipecat.processors.frame_processor import FrameProcessor, FrameDirection

class MyTranscriptHandler(FrameProcessor):
    async def process_frame(self, frame, direction):
        await super().process_frame(frame, direction)

        if isinstance(frame, TranscriptionFrame) and frame.user_id == "":
            print(f"Translated: {frame.text}")

        await self.push_frame(frame, direction)

---

Links

• Pinch
• Pipecat documentation
• Pipecat GitHub

---

License

Apache 2.0 — see LICENSE.