Function streamer

streamer.py:160–252

Process a batch of data items and generate audio, return audio data and metadata

(
    batch_items,
    tokenizer,
    model: AsteroidTTSInstruct,
    spt: XY_Tokenizer,
    device,
    system_prompt,
    use_normalize=False,
    use_tqdm=False,
)

def streamer(
    batch_items,
    tokenizer,
    model: AsteroidTTSInstruct,
    spt: XY_Tokenizer,
    device,
    system_prompt,
    use_normalize=False,
    use_tqdm=False,
):
    """Process a batch of data items and generate audio, return audio data and metadata"""
    # Prepare batch data
    batch_size = len(batch_items)
    assert batch_size == 1, "Batch size must be 1 for streaming"
    texts = []
    prompts = [system_prompt] * batch_size
    prompt_audios = []
    actual_texts_data = []  # Store actual text data used

    # Extract text and audio from each sample
    for i, item in enumerate(batch_items):
        # Use new processing function
        processed_item = process_jsonl_item(item)

        text = processed_item["text"]
        prompt_text = processed_item["prompt_text"]

        # Merge text, if prompt_text is empty, full_text is just text
        full_text = prompt_text + text if prompt_text else text
        original_full_text = full_text  # Save original text

        # Apply text normalization based on parameter
        if use_normalize:
            full_text = normalize_text(full_text)

        # Replace speaker tags
        final_text = full_text.replace("[S1]", "<speaker1>").replace(
            "[S2]", "<speaker2>"
        )
        texts.append(final_text)

        # Save actual text information used
        actual_texts_data.append(
            {
                "index": i,
                "original_text": original_full_text,
                "normalized_text": (
                    normalize_text(original_full_text) if use_normalize else None
                ),
                "final_text": final_text,
                "use_normalize": use_normalize,
            }
        )

        # Get reference audio
        prompt_audios.append(processed_item["prompt_audio"])

    # Process inputs