Transcription Fails for Files Without Extension in Multipart Encoder

[rezmoth]

Be reader of type io.Reader

transcription, err := client.Audio.Transcriptions.New(ctx, openai.AudioTranscriptionNewParams{
	Model: openai.F(openai.AudioModelWhisper1),
	File:  openai.F[io.Reader](reader),
})

The above call fails with:

POST "https://api.openai.com/v1/audio/transcriptions": 400 Bad Request {
  "error": {
    "message": "Unrecognized file format. Supported formats: ['flac', 'm4a', 'mp3', 'mp4', 'mpeg', 'mpga', 'oga', 'ogg', 'wav', 'webm']",
    "type": "invalid_request_error",
    "param": null,
    "code": null
  }
}

Edit:
Seems like filename in

openai-go/internal/apiform/encoder.go

Line 316 in 8d58dfa

filename := "anonymous_file"

requires a valid extension otherwise the call fails

Fix in PR #92 seems to fix it

[jacobzim-stl]

Thanks! I'll take a look

[jacobzim-stl]

The issue here is that the backend needs the file extension for decoding. If you need to process an anonymous file reader from, os.Stdin for instance, you'll have to manually add a filename to the reader.

package main

import (
	"context"
	"io"
	"os"

	"github.com/openai/openai-go"
)

type TranscriptionsReader struct {
	*os.File
}

func (TranscriptionsReader) Name() string {
	return "anonymous.mp3"
}

func main() {
	client := openai.NewClient()
	ctx := context.Background()

	var x TranscriptionsReader = TranscriptionsReader{os.Stdin}
	transcription, err := client.Audio.Transcriptions.New(ctx, openai.AudioTranscriptionNewParams{
		Model: openai.F(openai.AudioModelWhisper1),
		File:  openai.F[io.Reader](x),
	})

	if err != nil {
		panic(err)
	}

	println(transcription.Text)
}