How to return word level timestamp granularity from speech-to-text (whisper)?

[sohizi-entreprise]

Hello,

I want to transcribe an audio and return word granularity timestamp. This is supported by models like Openai Whisper-1.
However, even though I passed the correct parameters to the provider, OpenRouter only return the text.

async function transcribeAudio(audioBase64: string) {
    const response = await sttCreateTranscription(openRouter, {
        sttRequest: {
            inputAudio: {
                data: audioBase64,
                format: 'mp3',
            },
            model: 'openai/whisper-1',
            provider: {
                options: {
                    openai: {
                        timestamp_granularities: ["word"],
                        response_format: "verbose_json"
                    }
                }
            }
        }
    })
    if (!response.ok) {
        throw new Error(`Failed to transcribe audio: ${response.error.message}`);
    }

    return response.value;
}

I was expecting this to return an additional key "words" as the Openai SDK.