parlona 1.0.0

Call analytics pipeline for speech-to-text, summarization, and insights extraction

Parlona

Call analytics pipeline for speech-to-text, summarization, and insights extraction.

Installation

pip install parlona

Set your LLM API key (for OpenAI):

export OPENAI_API_KEY="your-api-key"

Quick Start

import parlona

# Process an audio file
result = parlona.process("call.wav")

# Access results
print(result.transcript)
print(result.summary)
print(result.headline)
print(f"Sentiment: {result.sentiment.label} ({result.sentiment.score})")
print(f"Entities: {result.entities}")

Features

• Speech-to-Text: Powered by faster-whisper with stereo channel diarization
• Call Summarization: LLM-powered summaries via OpenAI, Groq, vLLM, or Ollama
• Sentiment Analysis: Automatic sentiment detection and scoring
• Entity Extraction: Named entity recognition with speaker attribution
• Multi-language Support: Automatic language detection

Advanced Usage

from parlona import CallProcessor, STTConfig, LLMConfig

# Custom configuration
stt_config = STTConfig(
    model_name="Systran/faster-whisper-medium",
    device="cuda",
    diarization_mode="stereo_channels",
    speaker_mapping={0: "agent", 1: "customer"}
)

llm_config = LLMConfig(
    backend="openai",
    api_key="your-api-key",
    model="gpt-4o-mini"
)

# Create processor and process
processor = CallProcessor(stt_config=stt_config, llm_config=llm_config)
result = processor.process("call.wav")

Modular Usage

Use STT and LLM components separately:

from parlona.stt import STTEngine, STTConfig
from parlona.llm import LLMClient, LLMConfig

# STT only
stt_engine = STTEngine(STTConfig())
transcription = stt_engine.transcribe("audio.wav")

# LLM only
llm_client = LLMClient(LLMConfig(backend="openai"))
summary, headline, lang, sentiment, entities, score = \
    llm_client.summarize_with_headline(transcript)

Requirements

• Python 3.9+
• OpenAI API key (or other LLM backend)

License

Apache-2.0