Logo Lanfrica
  • Accueil
  • Atlas
  • Analyses
  • Documentation
  • Sign in

© 2026 Lanfrica. Tous droits réservés. Tous les droits d'auteur des ressources affichées sur le site Web Lanfrica appartiennent aux détenteurs de droits d'auteur d'origine, sauf indication contraire explicite.

PetrosGondwe/Chichewa-Acoustic-Data-Collector-Offline

Domaine:

natural language processing

Type de record:

software
Créateur:
Pet
Hôte:
Collect closed-set letter and syllable acoustic-recordings for standard 1 Chichewa book(SOSA). **Chichewa Acoustic Data Collection** is a Progressive Web App (PWA) designed for offline-first collection of acoustic audio data (letters and syllables) for standard 1 Chichewa (SOSA book). ### Key Components * **Frontend Single-Page Application (`index.html`)**: Uses raw HTML5, CSS3, and JavaScript along with embedded Dexie.js for IndexedDB management. * **Service Worker (`sw.js`)**: Manages offline caching and PWA app updates. * **Curriculum Configuration (`prompts.json`)**: Contains structured prompt targets (letters and syllables). * **ASR Integration Script (`merge_asr/merge_asr_datasets.py`)**: A Python utility to aggregate and process exported datasets. --- ### Core Functionality & Features * **Authentication & Security**: Features PBKDF2 hashing with salt, rate-limiting, and locked-out states after failed attempts. * **Audio Recording & Processing**: * Captures audio at 16kHz mono via standard Web MediaRecorder API. * Performs RMS volume checks and SNR/noise-floor rejection. * Auto-converts `.webm` audio to 16kHz 16-bit PCM `.wav` format upon zip export. * **Data Management**: * Block-based target workflow (47 prompts per block). * Direct file-system backup integration via File System Access API. * Export mechanism generating metadata CSV files and packaged ZIP archives. ### Field deployment and recovery Run the app from `https://` or `localhost`; microphone access, service-worker offline support, and folder backups are restricted by browser secure-context rules. Chrome or Edge on Android/desktop is the supported target for recording and File System Access backups. Keep the app open while a take is being saved; updates are deferred during recording and processing. Select a backup folder before collection and use **Backup Now** at handover. Also use **Export Dataset** regularly. A backup contains the database manifests and original WebM recordings; an export contains converted WAV files and `metadata.csv`. Do not edit bac …

Visit

github.com

Tasks

automatic speech recognitionspeech processing

Languages

ChichewaPhuie