Description #

This operator can be used for language identification modules. It is a multilingual LID (Language Identification**,** language identification) + ASR (Automatic Speech Recognition, automatic speech recognition) solution based on the Whisper model.

Key features #

• Multilingual recognition: Supports nearly one hundred languages including Chinese and English; also supports outputting language tags alongside the recognized text (for example, en, zh).
• Punctuation restoration: Customizable punctuation restoration for Chinese and English to improve the readability of the output text.
• Supports multiple audio input formats: URL, binary, and more

Supported models #

• Whisper series models (LID + ASR)
  • openai/whisper-large-v3-turbo
  • openai/whisper-large-v3
  • openai/whisper-medium (Chinese support is limited)
  • openai/whisper-small (may output traditional Chinese characters)
• Chinese and English punctuation restoration model
  • iic/punc_ct-transformer_cn-en-common-vocab471067-large

Language support #

Supports nearly one hundred languages including Chinese, English, German, Spanish, and more. See the complete language list. ‌

Recommended practices #

• Supports multiple audio input formats (URL, binary, and more)
• Batch processing of audio segments up to 30 seconds long
• Recognition accuracy is usually highest for English scenarios
• Punctuation restoration models can optionally be used for Chinese and English scenarios to improve text readability
• Supports returning only language identification results
• Supports GPU-accelerated inference; it is recommended to use CUDA devices with more than 4 GB of GPU memory

Operator parameters #

Input #

Output #

A structured result array, where each element is a struct containing the following fields:

• asr_result (str): The text result obtained from speech recognition.
• language (str): The recognized language code (for example, en, zh).
• asr_result_with_punc (Optional[str]): Optional text result with punctuation. Returned when the punctuation model is loaded and available during initialization; otherwise, None.

Parameters #

If a parameter does not have a default value, it is required.

Examples #

The following code demonstrates how to use daft to run this operator for speech language identification and transcription.

from __future__ import annotations

import logging
import os

import ray

import daft
from daft import col
from daft.las.functions.audio.audio_asr_lid_whisper import AudioAsrLidWhisper
from daft.las.functions.udf import las_udf

if __name__ == "__main__":
    if os.getenv("DAFT_RUNNER", "ray") == "ray":

        def configure_logging():
            logging.basicConfig(
                level=logging.INFO,
                format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
                datefmt="%Y-%m-%d %H:%M:%S.%s".format(),
            )
            logging.getLogger("tracing.span").setLevel(logging.WARNING)
            logging.getLogger("daft_io.stats").setLevel(logging.WARNING)
            logging.getLogger("DaftStatisticsManager").setLevel(logging.WARNING)
            logging.getLogger("DaftFlotillaScheduler").setLevel(logging.WARNING)
            logging.getLogger("DaftFlotillaDispatcher").setLevel(logging.WARNING)

        import ray

        ray.init(dashboard_host="0.0.0.0", runtime_env={"worker_process_setup_hook": configure_logging})
        daft.set_runner_ray()

    daft.set_execution_config(actor_udf_ready_timeout=600)
    daft.set_execution_config(min_cpu_per_task=0)

    tos_dir_url = os.getenv("TOS_DIR_URL", "las-cn-beijing-public-online.tos-cn-beijing.volces.com")
    samples = {
        "audio_path": [
            f"https://{tos_dir_url}/public/shared_audio_dataset/sample_normal.wav"
        ]
    }

    model_path = os.getenv("MODEL_PATH", "/opt/las/models")
    model_name = "openai/whisper-large-v3"
    audio_src_type = "audio_url"
    punc_model_name = "iic/punc_ct-transformer_cn-en-common-vocab471067-large"
    num_gpus = 1
    device = "cuda" if num_gpus > 0 else "cpu"
    batch_size = 1
    return_language_only = False

    df = daft.from_pydict(samples)
    df = df.with_column(
        "asr_result_detail",
        las_udf(
            AudioAsrLidWhisper,
            construct_args={
                "audio_src_type": audio_src_type,
                "model_path": model_path,
                "model_name": model_name,
                "punc_model_name": punc_model_name,
                "return_language_only": return_language_only,
                "batch_size": batch_size,
                "device": device,
            },
            num_gpus=num_gpus,
            batch_size=1,
            num_cpus=4,
            concurrency=1,
        )(col("audio_path")),
    )

    df.show()

    # ╭───────────────────┬──────────────────────────────────────────╮
    # │ audio_path                     ┆ asr_result_detail                                                    │
    # │ ---                            ┆ ---                                                                  │
    # │ Utf8                           ┆ Struct[asr_result: Utf8, language: Utf8, asr_result_with_punc: Utf8] │
    # ╞═══════════════════╪══════════════════════════════════════════╡
    # │ https://las-cn-beijing-publi-… ┆ {asr_result: 人我保住了金我取到了俺老孙啥功名…                            │
    # ╰───────────────────┴──────────────────────────────────────────╯