Replies: 2 comments
-
PromptThe prompt is not used to provide "commands" for the model but rather to give context for the audio and improve the accuracy and grammar of the final transcript. TemperatureTemperature controls how "creative" the speech recognition is, using more common words at low values and more surprising (and less common) words at high values, with possible values between refs: openai.com | ai.stackexchange.com | https://github.com/ggerganov/whisper.cpp ThreadsHow many cpu threads will the whisper engine run concurrently. It must be at least one. If you choose a higher value, it should decode faster but will use more resources (and the pc can heat up more, etc.) |
Beta Was this translation helpful? Give feedback.
-
Add more info in the latest version |
Beta Was this translation helpful? Give feedback.
-
Is it like a prompt used in LLM? I've wrote in the prompt area a prompt related to the transcribed text (i.e In Hebrew recording related to the stock market to replace words in Hebrew to English and to add the stock ticker. For example, אפל to Apple [AAPL]). Also asked to get a summary of the transcription and action items. It didn't do anything.
Also, how to use "threads" option? My understanding the that "temperature" should be kept at 0 when transcribing.
Thank you for this amazing tool.
Regards,
Isaac
Beta Was this translation helpful? Give feedback.
All reactions