redenlab-extract 1.3.0

Python SDK for RedenLab ML inference service

RedenLab Extract SDK

Python SDK for RedenLab's ML inference service. Run machine learning models on audio files using RedenLab's proprietary endpoints.

Features

• JWT-based authentication — browser login for humans, env vars for machines
• Simple session + client pattern: upload once, run any model
• Async job management with polling and progress callbacks
• Automatic chunking for long audio files
• Built-in retry logic and error handling

Installation

pip install redenlab-extract

Authentication

The SDK uses JWT tokens issued by AWS Cognito. Two modes are auto-detected:

Human / Interactive (PKCE)

Run once from the terminal:

redenlab login

This opens a browser, authenticates you, and stores tokens in ~/.redenlab-ml/credentials.yaml. The SDK refreshes tokens silently — you only need to run redenlab login again if your session expires.

Machine / Headless (Client Credentials)

Set environment variables before running your script:

export REDENLAB_CLIENT_ID=your-client-id
export REDENLAB_CLIENT_SECRET=your-client-secret

The SDK detects these and fetches a token automatically. No browser, no disk tokens.

Priority: if both env vars are set, Client Credentials is used regardless of any stored PKCE tokens.

Quick Start

from redenlab_extract import RedenLabSession, intelligibility_client, run_inference

session = RedenLabSession()
client = intelligibility_client(session, model_name="als-intelligibility-v1")

result = run_inference(session, client, "audio.wav")
print(result["result"]["intelligibility_score"])

from redenlab_extract import RedenLabSession, TranscribeClient, run_inference

session = RedenLabSession()
client = TranscribeClient(session, language_code="en-US")

result = run_inference(session, client, "audio.wav")
print(result["transcript"])

Available Models

Intelligibility

from redenlab_extract import intelligibility_client

client = intelligibility_client(session, model_name="als-intelligibility-v1")
# model_name options: "als-intelligibility-v1", "als-intelligibility-v2",
#                     "ataxia-intelligibility"

Naturalness

from redenlab_extract import naturalness_client

client = naturalness_client(session, model_name="als-naturalness-v1")
# model_name options: "als-naturalness-v1", "als-naturalness-v2",
#                     "ataxia-naturalness"

Transcription (AWS Transcribe)

from redenlab_extract import TranscribeClient

client = TranscribeClient(
    session,
    language_code="en-US",   # optional, defaults to "en-US"
    max_speaker_count=4,      # optional, defaults to 10
)

WhisperX

from redenlab_extract import whisperX_client

client = whisperX_client(session, diarize=True, batch_size=16)

Speaker Counter

from redenlab_extract import speaker_counter_client

client = speaker_counter_client(session)

Speaker Diarization

from redenlab_extract import speaker_diarization_client

client = speaker_diarization_client(session)

Sylber Time

from redenlab_extract import sylber_time_client

client = sylber_time_client(session)

Usage Patterns

Level 1 — run_inference() (recommended)

Handles upload, chunking, and inference in one call:

from redenlab_extract import RedenLabSession, intelligibility_client, run_inference

session = RedenLabSession()
client = intelligibility_client(session, model_name="als-intelligibility-v1")

result = run_inference(session, client, "audio.wav")

Long files are chunked automatically and results are aggregated.

Level 2 — Upload → Submit → Poll

Useful for batch processing or when you want to decouple upload from inference:

# Upload
asset = session.upload("audio.wav")

# Submit (with retry for async S3 processing)
job_id = client._submit_with_upload_wait(asset)

# Poll
result = client.poll(job_id, timeout=600)

Level 3 — Batch processing

Submit all jobs first, then poll concurrently:

# Upload and submit all files
assets = [session.upload(f) for f in audio_files]
job_ids = [client._submit_with_upload_wait(asset) for asset in assets]

# Poll each
results = [client.poll(job_id) for job_id in job_ids]

Progress callbacks

def on_progress(status):
    print(f"Status: {status['status']}")

result = run_inference(session, client, "audio.wav", progress_callback=on_progress)

Chunked inference (long files)

run_inference() chunks automatically. For manual control:

from redenlab_extract import run_chunked_inference, aggregate_results

chunks = run_chunked_inference(session, client, "long_audio.wav", chunk_duration=50)
result = aggregate_results(chunks, score_key="intelligibility_score")

Backend warmup

SageMaker-backed models (intelligibility, naturalness) may be cold after inactivity:

health = client.health()
if not health["ready"]:
    client.warmup()  # submits a dummy job to wake the endpoint

Configuration

Base URL

Resolved in priority order:

Source	How
Constructor	RedenLabSession(base_url="https://...")
Environment variable	REDENLAB_ML_BASE_URL=https://...
Config file	~/.redenlab-ml/config.yaml → base_url: ...
Service discovery	Automatic fallback

Config file (~/.redenlab-ml/config.yaml)

base_url: https://your-api-gateway-url.amazonaws.com/prod  # optional

Credentials file (~/.redenlab-ml/credentials.yaml)

Written automatically by redenlab login. Do not edit manually.

Supported File Types

• WAV (.wav)
• FLAC (.flac)
• MP3 (.mp3)
• OGG (.ogg)
• M4A (.m4a)

Requirements

• Python 3.10+
• requests
• tenacity
• pyyaml

Optional (for audio chunking):

• soundfile
• numpy

Support

• Email: [e-mail]
• Website: https://redenlab.com

License

MIT License — see LICENSE for details.

Changelog

See CHANGELOG.md for version history.