transcribe-diarize-emote 0.1.1

End-to-end audio pipeline: transcription + diarization + emotion enrichment

transcribe-diarize-emote

One-stop audio intelligence pipeline combining:

• Transcription (faster-whisper)
• Speaker Diarization (diarizer-lite)
• Acoustic Emotion Context Enrichment (whisper-speaker-emotion)

Designed to produce LLM-ready conversational transcripts in a single pass.

What This Library Solves

Modern STT systems give you text.
But real conversation intelligence requires:

• A clean transcription system
• Who spoke → Diarization
• How was it said → Acoustic emotion context
• Clean speaker turns → LLM context engineering

transcribe-diarize-emote solves all three in one pipeline.

Why This Matters For LLM Workflows

Output can be directly used for:

• Summarization
• Sentiment Analysis
• Topic Modeling
• Call Quality Monitoring
• Compliance Analytics
• Customer Experience Intelligence
• Conversation Mining
• Agent Coaching
• Risk / Escalation Detection

Key Features

• Offline capable
• No API keys required
• CPU supported by default
• CUDA GPU acceleration
• Emotion layer optional toggle
• Direct LLM-ready outputs

Installation

pip install transcribe-diarize-emote

Core Usage

from transcribe_diarize_emote import Pipeline

pipe = Pipeline(
    emotion=True,
    whisper_model="large-v3",
    device="cpu",
    compute_type="int8"
)

result = pipe.process("call.wav")

print(result.vtt)

Pipeline Configuration Explained

Emotion Layer (emotion)

Setting	Meaning
emotion=True	Adds acoustic emotion metrics
emotion=False	Faster pipeline, text only

Whisper Model (whisper_model)

Recommended

• large-v3

Other Supported Models

• tiny
• base
• small
• medium
• large-v2
• large-v3 (recommended for quality)

Device Support (device)

Device	Supported
CPU	Yes
CUDA GPU	Yes
Apple MPS	No

Compute Type (compute_type)

Device	Recommended
CPU	int8
CUDA	float16

Audio Format

The major supported audio formats are:

• .mp3
• .wav

Output Behavior

emotion=False (Default)

Recommended Outputs

• result.text
• result.json
• result.vtt

Sample Json

[
  {
    "start": 0.0,
    "end": 4.0,
    "speaker": "Speaker0",
    "text": "Hi, I wanted to report an issue with my ride this morning."
  },
  {
    "start": 4.0,
    "end": 7.0,
    "speaker": "Speaker1",
    "text": "Sure, could you tell me what happened?"
  },
  {
    "start": 7.0,
    "end": 13.0,
    "speaker": "Speaker0",
    "text": "Yeah, driver was polite but car wasn't clean and smelled weird."
  },
  {
    "start": 13.0,
    "end": 17.0,
    "speaker": "Speaker1",
    "text": "I'm sorry to hear that. Anything else?"
  },
  {
    "start": 17.0,
    "end": 25.0,
    "speaker": "Speaker0",
    "text": "Also he took a longer route even after I gave the correct address. Got it. We will look into this and report back."
  }
]

emotion=True

Recommended Output

• result.vtt (Better for readability and LLM understanding)

Sample vtt

WEBVTT

Average Pitch:
Speaker0: 230.45 Hz
Speaker1: 190.78 Hz

Average RMS:
Speaker0: 0.068
Speaker1: 0.061

00:00:00.000 --> 00:00:04.000
<v Speaker0> Hi, I just wanted to give some feedback about my cab ride this morning.
Pitch: 236.22 Hz | RMS: 0.066
Pitch Difference with Average: +5.77 Hz
RMS Difference with Average: -0.002

00:00:04.000 --> 00:00:07.000
<v Speaker1> Sure, I can help you with that. Could you tell me what happened?
Pitch: 192.45 Hz | RMS: 0.062
Pitch Difference with Average: +1.67 Hz
RMS Difference with Average: +0.001

Output Philosophy

emotion=False

Structured conversational transcript optimized for:

• Summarization
• Classification
• Topic modeling
• Search indexing

emotion=True

Emotion-enriched transcript including (Additional context for LLM for sentiment analysis):

• Speaker baseline voice metrics
• Pitch (F0)
• Loudness (RMS)
• Emotional deviation from baseline tone

Citations

This library builds on the following open-source technologies:

- faster-whisper

High-performance Whisper inference powered by **CTranslate2**, enabling fast CPU and GPU transcription without requiring PyTorch runtime.
Project: https://pypi.org/project/faster-whisper/

- diarizer-lite

Lightweight stereo energy-based speaker diarization designed for offline processing and post-STT conversational structuring.
Project: https://pypi.org/project/diarizer-lite/

- whisper-speaker-emotion

Emotion-aware speaker analysis using audio pitch and loudness aligned with Whisper diarized VTT transcripts.
Project: https://pypi.org/project/whisper-speaker-emotion/

Designed For LLM Context Engineering

Traditional LLM pipelines rely purely on text transcripts.

However, real-world conversations contain acoustic emotional signals that text alone cannot capture.

Instead of forcing LLMs to infer emotion from text,
transcribe-diarize-emote provides measured acoustic context directly from audio.

This enables LLM systems to:

• Detect emotional spikes
• Improve sentiment confidence scoring
• Identify escalation or frustration moments
• Improve summarization quality
• Improve explainability of sentiment and classification outputs

This is an example of context engineering — improving LLM outputs by improving the quality and richness of input context rather than modifying or tuning the model itself.

Contributing

Pull requests are welcome.

For major changes, please open an issue first to discuss what you would like to change.

Please ensure tests and documentation are updated where applicable.

License

MIT License