By default OpenAI returns an mp3 while Kokoros seems to return a wav

[rikhuijzer]

From the OpenAI docs, OpenAI returns an mp3:

curl https://api.openai.com/v1/audio/speech \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tts-1",
    "input": "Today is a wonderful day to build something people love!",
    "voice": "alloy"
  }' \
  --output speech.mp3

But if I look at the file that I get from the Kokoros OpenAI compatible endpoint, it looks like it's a wav file?

$ file tests/tmp-openai-compatible.mp3
tests/tmp-openai-compatible.mp3: RIFF (little-endian) data, WAVE audio, IEEE Float, mono 24000 Hz

This is the request I made

$ export HOST="http://localhost:3000"

$ curl "$HOST/v1/audio/speech" \
  --verbose \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tts-1",
    "input": "Today is a wonderful day to build something people love!",
    "voice": "am_adam"
  }' \
  --output speech.mp3

VLC is unable to play it and ffmpeg also crashes when trying to convert it (while it worked with OpenAI's mp3). The vscode audio-preview extension can play it without problems.

[lucasjinreal]

Hello, this format should be able to modify when user request, wav is the default format used in save, would u like add a PR to support mp3 return specific in openai returning?

[7jrxt42BxFZo4iAnN4CX]

#68

curl -X POST http://localhost:3000/v1/audio/speech \
                -H "Content-Type: application/json" \
                -d '{
              "model": "anything can go here",
              "input": "Hello, this is a test of the Kokoro TTS system!",
              "voice": "af_sky", 
              "response_format": "mp3"
            }' --output sky-says-hello.mp3

[gnusupport]

ffmpeg -i "${output_file}" "${m4a}" 2>&1

I'm normally converting the mp3 to m4a file so that it is ready to paste or share on chat systems and then people can easily listen to the voice message.