Failed to load model: Failed to create a new whisper context

[TobseF]

Bug Description

Bug:

Loading the downloaded Whisper ++ Large v3 Turbo model fails with:

Failed to load model: Failed to create a new whisper context

[
  {
    "error": {
      "name": "WhisperingError",
      "severity": "error",
      "title": "🤖 Model Loading Error",
      "description": "Failed to load model: Failed to create a new whisper context.",
      "action": {
        "type": "more-details",
        "error": {}
      }
    },
    "data": null
  }
]

Expected behavior:

Audio should be transcribed into text.

Steps to reproduce:

1. Record audio
2. Click on "Transcribe selected recordings"
3. See error message

Additional context:

Windows 11 de x64 GPU: RTX 4070 TI Super 16 GB VRAM Model: Whisper ++ Large v3 Turbo Audio Compression: Disabled Language: German Whispering v7.5.4

Platform

Mac OS

Affected Component

Transcription

Whispering Version

7.5.4

Severity

Moderate... Only cloud options are available

Discord Link

No response

Checklist

• [x] I have searched existing issues to ensure this hasn't been reported
• [x] I have included all relevant information above

[ch3rub1n]

Same here on Windows 11, app version 7.5.4, tested on 7.5.2 as well with same result

Whisper Model: Small Audio Compression: Disabled

Failed to load model: Failed to create a new whisper context.

@TobseF You mentioned in your report Severity: Moderate (workaround available)

Can you please share what is the workaround you use?

[TobseF]

@ch3rub1n Sorry, I have no workaround. I just missed that point in the issue template 😅 Thanks for the hint. I changed the text to "Only cloud options are available".

[DefaultPerson]

Same problem, try to download the model manually, the built-in downloader may not download the high models completely. It helped me

[ch3rub1n]

Same problem, try to download the model manually, the built-in downloader may not download the high models completely. It helped me

@DefaultPerson Bingo! Thank you.

I downloaded the model from https://huggingface.co/ggerganov/whisper.cpp/tree/main ggml-small.bin

The file I had in my folder, one that was downloaded by the app was 362MB

The one I downloaded manually is 465MB and it works.

Does anyone know if it's possible to get rid of the command prompt window (I assume that's what it is) flash when transcription starts?

[TobseF]

⬇️ Thanks for the hint. I used the the ggml-large-v3-turbo and it had only 600MB instead of 1.51 GB. I download it here: https://huggingface.co/ggerganov/whisper.cpp/blob/main/ggml-large-v3-turbo.bin and copied it to C:\Users\tobse\AppData\Roaming\com.bradenwong.whispering\whisper-models. Now it works ✅