unknown

A developer at MaibornWolff tested OpenAI's Whisper AI on a Raspberry Pi 4 to evaluate its live transcription capabilities on a low-power device. Whisper, a versatile speech recognition system, excels at understanding spoken words, translating languages, and identifying speakers. Despite the Raspberry Pi's limited processing power, the experiment aimed to see if live transcription could still be effective. Using projects like Whisper-Ctranslate2 and Faster-Whisper, the developer faced challenges like slow reaction times and issues with word detection. Adjustments in the code and leveraging specific Python libraries helped address these problems, making live transcription possible though with certain limitations. While the Tiny and Base models were found suitable for live transcription, higher accuracy models were too slow for real-time use.

OpenAI Whisper auf dem Raspberry Pi 4 – Live-Transkription

This site if for sharing nuggets only. It is mainly focusing on sites that touch the domain of learning. It is opening for positing for everyone at the moment, but this might change, depending on the amount of sh1t posted.

Maschinenpriester

How to run OpenAI's Whisper on a Raspberry Pi 4 (or 5) with 8GB of RAM. The model was facing three problems according to the author:
1. Slow reaction speed when transcribing. 
2. Mangled up responses, when a second recording is being transmitted, while the first is till being processed.
3. The first word of the transcription is often missing, due to the way Whisper is recognizing speech.

The author is making changes to the code to address those problems. In the end, a really in-depth article.

<p>How to run OpenAI's Whisper on a Raspberry Pi 4 (or 5) with 8GB of RAM. The model was facing three problems according to the author:
1. Slow reaction speed when transcribing. 
2. Mangled up responses, when a second recording is being transmitted, while the first is till being processed.
3. The first word of the transcription is often missing, due to the way Whisper is recognizing speech.

The author is making changes to the code to address those problems. In the end, a really in-depth article. </p>