Podonos

Founded 2024

Evaluate and benchmark voice AI models across multiple languages and accents

Pricing model

Paid

You can use Podonos to test, evaluate, and benchmark voice AI and speech synthesis models. It identifies audio issues such as mispronunciations, fuzzy sound, audio distortion, and omitted words across multiple languages and locales. You can run customized human evaluation studies with pre-screened participants using standards like ITU-R P.835 or automate evaluations via the Python library. It provides interactive reports, preference testing between text-to-speech models, and detailed quality metrics to guide engineering decisions before releasing voice models to production.

What Podonos does

Run speech synthesis preference testsCollect Mean Opinion Scores for synthetic audioScreen audio evaluators for hearing accuracy and focusBenchmark text-to-speech models using PythonIdentify distortion and dropped words in generated voicesStandardize audio playback levels across test setsShare voice evaluation reports with team membersMultilingual evaluations across 9+ languages and 13 localesPre-screening with device, attention, intelligibility, and noise detection testsPython library for automated evaluation pipelinesITU-R P.835 standard compliance for speech evaluationsSide-by-side TTS model preference testingAutomated audio level standardizationInteractive visual reporting dashboards