Tyto
Use audio risk and quality indicators to investigate downstream voice AI failures.
Tyto is an audio insight model by ai-coustics. It predicts whether the audio reaching your agent will cause downstream failures, with a single risk score and the likely cause behind it. Catch degraded audio before it breaks your voice agent β in real time or post-call. Real-time audio intelligence that makes Voice AI work in production. Not just in the lab. Inference runs through the aic-sdk Python package (also available for Rust, C/C++, Node.js, and WebAssembly).
Setup
- Install the SDK:
pip install aic-sdk - Create a free SDK key on the ai-coustics playground β 30-day trial included, no credit card required.
Usage
# initialize the audio insight analyzer
import aic_sdk as aic
model = aic.Model.from_file(aic.Model.download("tyto-1.2-l-16khz", "./models"))
analyzer = aic.FileAnalyzer(model, "{aic-sdk-key}") # get a free SDK key: https://developers.ai-coustics.com/login
# score a mono float32 buffer β one AnalysisResult per 5-second window
results = analyzer.analyze(audio, sample_rate)
# predict downstream failures, with the exact cause behind them
for window in results:
print(f"risk={window.risk_score:.2f} noise={window.noise:.2f} interfering={window.interfering_speech:.2f}")
What you get
Every analysis window returns one headline score plus six diagnostic dimensions (all 0.0β1.0; lower is less problematic, except loudness):
| Field | Meaning |
|---|---|
risk_score |
Headline score β predicted likelihood of failure of downstream STT, VAD, turn-taking, or speech-to-speech models |
noise |
Ambient / environmental noise |
speaker_reverb |
Speaker distance and reverberation |
speaker_loudness |
Speaker loudness |
interfering_speech |
Interference from additional speakers |
packet_loss |
Dropouts / discontinuities (packet loss, frame erasure, jitter, CPU overload) |
codec_degradation |
Artifacts from lossy or narrowband speech codecs |
Streaming / real-time analysis
For live audio, use a Collector/Analyzer pair: the collector buffers on the audio thread (real-time safe), the analyzer runs on a separate thread (analysis is computationally expensive):
import aic_sdk as aic
model = aic.Model.from_file(aic.Model.download("tyto-1.2-l-16khz", "./models"))
collector, analyzer = aic.analyzer_pair(model, "{aic-sdk-key}") # get a free SDK key: https://developers.ai-coustics.com/login
config = aic.ProcessorConfig.optimal(model)
collector.initialize(config)
# audio thread (real-time safe):
for block in audio_blocks(): # np.ndarray, config.block_size samples
collector.buffer(block)
# non-audio thread, whenever you want a reading:
result = analyzer.analyze_buffered()
print(result.risk_score, result.packet_loss, result.codec_degradation)
Related ai-coustics models
- π Quail Voice Focus β primary-speaker isolation for speech-to-text and voice agents
- π Quail Multi Speaker β multi-speaker speech enhancement for speech-to-text
- π§ Rook Multi Speaker β speech enhancement for human listening
- ποΈ VAD Multi Speaker β noise-robust VAD for any speaker
- ποΈ VAD Voice Focus β primary-speaker VAD for voice agents
Resources
- π Tyto docs Β· SDK quickstart Β· docs.ai-coustics.com
- π§© Framework quickstarts: LiveKit, Pipecat
- ποΈ All model versions & checksums: artifacts.ai-coustics.io
- π¦ Python SDK: pypi.org/project/aic-sdk
- π Free SDK key: developers.ai-coustics.com
- π’ Enterprise licensing: ai-coustics.com
- Downloads last month
- 660