LangChain

langchain-prosodyai gives a LangChain agent one standard tool for recorded audio. The result is the complete structured ProsodyAI response: transcript, recording-local speaker turns, ordered acoustic measurements, and same-speaker deltas.

Install from the public main branch:

python -m pip install \
  "langchain-prosodyai @ git+https://github.com/ProsodyAI/langchain.git@main"

import os
from pathlib import Path

from langchain_prosodyai import ProsodyAnalyzeAudioTool

tool = ProsodyAnalyzeAudioTool(
    api_key=os.environ["PROSODY_API_KEY"],
    allowed_audio_root=Path("./recordings"),
)

analysis = tool.invoke({"audio_path": "recording.wav", "language": "en"})

print(analysis["text"])

for turn in analysis.get("turns") or []:
    print(turn["speaker_id"], turn["start_ms"], turn["end_ms"], turn["text"])

for frame in analysis.get("prosody_timeline") or []:
    print(frame["speaker_id"], frame.get("acoustic_state"))
    print(frame.get("acoustic_change"))

The model receives a path relative to allowed_audio_root; it never receives unrestricted filesystem access. Supported extensions are .wav, .mp3, .m4a, .flac, and .ogg.

The tool returns the recorded-audio API object unchanged, with every turn and every measured frame intact. A speaker_id is scoped to a single recording.

Pass the same tool to create_agent when a model should be able to request an analysis:

from langchain.agents import create_agent

agent = create_agent(
    model="openai:gpt-5.4",
    tools=[tool],
)

See the public integration repository for the direct client, compatibility policy, and release status.

Guides

LangChain

On this page