描述 #

语音识别模块 - 基于Whisper模型的多语言语音转文字解决方案

核心功能 #

• 多语言识别：支持中英文等主流语言
• 语音翻译：可将识别结果翻译为英文

推荐实践 #

• 30秒内长度的音频片段可支持 batch_size > 1，提高处理效率；否则建议 batch_size 设置为1
• 英文场景识别准确率最高

支持模型 #

• openai/whisper-large-v3-turbo
• openai/whisper-large-v3
• openai/whisper-medium（中文支持一般）
• openai/whisper-small（中文输出繁体字）

语种支持 #

完整语种列表请参考：
https://github.com/ggml-org/whisper.cpp/blob/d682e150908e10caa4c15883c633d7902d385237/src/whisper.cpp#L248

算子参数 #

输入 #

输出 #

处理后的结构化数组，每个元素包含以下字段：

• asr_result: 语音识别文本结果；
• timestamps: 时间戳对列表(开始/结束时间)；
• segments: 分段文本结果列表

参数 #

如参数没有默认值，则为必填参数

调用示例 #

下面的代码展示了如何使用 daft 运行算子将语音转换为文字。

from __future__ import annotations

import logging
import os

import ray

import daft
from daft import col
from daft.las.functions.audio.audio_asr_whisper import AudioAsrWhisper
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.context.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/参观八达岭长城。.wav"
        ]
    }

    model_path = os.getenv("MODEL_PATH", "/opt/las/models")
    model_name = "openai/whisper-large-v3"
    audio_src_type = "audio_url"
    dtype = "bfloat16"
    source_language = "chinese"
    translate_to_english = False
    condition_on_prev_tokens = True
    compression_ratio_threshold = 1.35
    temperature = 0.5
    logprob_threshold = -1.0
    batch_size = 1
    rank = 0

    df = daft.from_pydict(samples)
    df = df.with_column(
        "asr_result_detail",
        las_udf(
            AudioAsrWhisper,
            construct_args={
                "audio_src_type": audio_src_type,
                "model_path": model_path,
                "model_name": model_name,
                "dtype": dtype,
                "source_language": source_language,
                "translate_to_english": translate_to_english,
                "condition_on_prev_tokens": condition_on_prev_tokens,
                "compression_ratio_threshold": compression_ratio_threshold,
                "temperature": temperature,
                "logprob_threshold": logprob_threshold,
                "batch_size": batch_size,
                "rank": rank,
            },
            num_gpus=1,
            batch_size=1,
            concurrency=1,
        )(col("audio_path")),
    )
    df.show()

    # ╭────────────────────────────────┬─────────────────────────────────────────────────────────────╮
    # │ audio_path                     ┆ asr_result_detail                                           │
    # │ ---                            ┆ ---                                                         │
    # │ Utf8                           ┆ Struct[asr_result: Utf8, timestamps: List[List[Float32]],   │
    # │                                ┆ segments: List[Utf8]]                                       │
    # ╞════════════════════════════════╪═════════════════════════════════════════════════════════════╡
    # │ tos://las-cn-beijing-publi-…   ┆ {asr_result: 参观八道岭长城,                                  │
    # │                                ┆ timesta…                                                    │
    # ╰────────────────────────────────┴─────────────────────────────────────────────────────────────╯