Spanish Audio Transcription Software(English, Spanish, French) for Brand Detection in Radio Recording
Budget: $50 – $250 USD
Project Overview:
We are seeking to develop a robust and highly accurate audio-to-text transcription solution specifically designed for Spanish audio recordings (English, French), particularly from radio broadcasts. The primary focus is on detecting brand mentions and keywords with an accuracy of 95% or higher across various Latin American accents.
* **Language & Accuracy:** Spanish (Latin American variants), English, French. Minimum 90% accuracy across diverse accents, handling noisy audio (music, jingles, overlapping voices). Proper punctuation and time-stamping (ideal). Speaker labeling (optional).
* **Audio Input & Processing:** MP3/WAV files (30s–10 min each). Batch processing for hundreds to thousands of files monthly.
* **Keyword Detection:** Custom dictionary of brand names/slogans. Output in JSON/CSV: audio filename, detected brand, number of mentions. Timestamps/excerpts (optional).
* **Deployment:** Custom-trained open-source models (Whisper, Vosk, Silero, etc.) preferred. Self-hosted or Docker-based. Web UI for upload/review (optional). Avoid reliance on commercial APIs.
**Bonus Skills:** Fine-tuning speech models for Spanish (noisy data), French, English, audio preprocessing, keyword spotting/NLP, OpenAI
**Deliverables:** Working prototype (95%+ accuracy), brand/keyword detection engine, batch processing tool/script, Docker deployment documentation, basic dashboard (optional).
We are seeking to develop a robust and highly accurate audio-to-text transcription solution specifically designed for Spanish audio recordings (English, French), particularly from radio broadcasts. The primary focus is on detecting brand mentions and keywords with an accuracy of 95% or higher across various Latin American accents.
* **Language & Accuracy:** Spanish (Latin American variants), English, French. Minimum 90% accuracy across diverse accents, handling noisy audio (music, jingles, overlapping voices). Proper punctuation and time-stamping (ideal). Speaker labeling (optional).
* **Audio Input & Processing:** MP3/WAV files (30s–10 min each). Batch processing for hundreds to thousands of files monthly.
* **Keyword Detection:** Custom dictionary of brand names/slogans. Output in JSON/CSV: audio filename, detected brand, number of mentions. Timestamps/excerpts (optional).
* **Deployment:** Custom-trained open-source models (Whisper, Vosk, Silero, etc.) preferred. Self-hosted or Docker-based. Web UI for upload/review (optional). Avoid reliance on commercial APIs.
**Bonus Skills:** Fine-tuning speech models for Spanish (noisy data), French, English, audio preprocessing, keyword spotting/NLP, OpenAI
**Deliverables:** Working prototype (95%+ accuracy), brand/keyword detection engine, batch processing tool/script, Docker deployment documentation, basic dashboard (optional).