Whisper
OpenAI's speech-recognition project for transcription and translation into English.
- Multilingual transcription and language identification
- Local model weights and Python code
- Model sizes for different memory and speed requirements
Whisper is a speech-recognition model and software project from OpenAI . The repository includes weights and code for transcription, language identification, and translation into English. It can run locally instead of sending recordings to a hosted transcription service.
Running Whisper
The code and weights are MIT-licensed. Local operation uses the available processor, memory, and storage; rented hardware adds its own charges. Hosted transcription APIs are separate services with their own pricing and file limits. Review important transcript passages against the original recording, especially names, numbers, and unclear speech.
Hosted alternatives
Local Whisper keeps inference on the selected machine and requires maintaining its hardware and software. A hosted transcription service handles model serving and charges for audio or session usage.
- Deepgram offers recorded and streaming transcription alongside speech generation.
- AssemblyAI offers transcription and audio analysis; its streaming bill includes idle connection time.
- ElevenLabs combines transcription with speech and voice-agent products.
Choose according to the recording format, languages, live-streaming needs, and permitted handling of the audio.
Sources
Checked .