Download and install the APK:
The backend is deployed on Hugging Face Spaces and can be tested via the FastAPI documentation:
https://makss7-voice-analyzer-api.hf.space/docs
Voice Analyzer is a tool made for analyzing and evaluating human speech recordings. It includes a mobile application built with Expo Go (React Native + TypeScript) and a Python backend API that processes audio, extracts features, and returns analysis results.
The project consists of two main components:
The mobile app provides the user interface and is developed using TypeScript and React Native with Expo Go. It allows users to record or upload audio and view analysis results.
The app has two pages:
- Analyze audio – allows to extract linguistic and speech characteristics
- Evaluate audio – assesses overall audio quality
On both pages, users can:
- record audio directly from the device microphone
- upload existing audio files
- play back audio after processing
- view results returned by the backend
The app communicates with the backend through REST API requests.
The backend is implemented in Python and includes two endpoints.
/analyze– performs transcription and speech analysis/evaluate– computes acoustic quality metrics and overall score of the audio
Used for extraction of speech and language information when /analyze endpoint receives a request. This module uses OpenAI Whisper for speech recognition.
It provides:
- Speech transcription
- Language detection
- Word count
- Words per minute (WPM)
- Sound purity classification
Used to compute acoustic quality metrics from the audio signal with Praat/Parselmouth when /evaluate endpoint receives a request.
It extracts:
- Jitter – pitch period variability
- Shimmer – amplitude variability
- HNR – harmonics-to-noise ratio
- SNR – signal-to-noise ratio
An overall audio quality score is calculated based on those metrics.
Frontend
- React Native
- Expo Go
- TypeScript
Backend
- Python
- FastAPI / Flask
- OpenAI Whisper
- The user records or uploads an audio file in the mobile app.
- The file is sent to the Python backend via HTTP request.
- The backend:
- converts audio to WAV
- transcribes speech using Whisper
- extracts acoustic features using Parselmouth
- The backend returns structured JSON data.
- The mobile app displays the results:
- Analyze page shows transcription and speech metrics
- Evaluate page shows acoustic metrics and quality score