Alibaba Cloud Model Studio：SDK Python untuk Pengenalan Ucapan Real-time Paraformer

Mengenali ucapan dari mikrofon

import os
import signal  # untuk penanganan acara keyboard (tekan "Ctrl+C" untuk menghentikan perekaman)
import sys

import dashscope
import pyaudio
from dashscope.audio.asr import *

mic = None
stream = None

# Atur parameter rekaman
sample_rate = 16000  # laju pengambilan sampel (Hz)
channels = 1  # saluran mono
dtype = 'int16'  # tipe data
format_pcm = 'pcm'  # format data audio
block_size = 3200  # jumlah frame per buffer


def init_dashscope_api_key():
    """
        Atur DashScope API-key Anda. Informasi lebih lanjut:
        https://github.com/aliyun/alibabacloud-bailian-speech-demo/blob/master/PREREQUISITES.md
    """

    if 'DASHSCOPE_API_KEY' in os.environ:
        dashscope.api_key = os.environ[
            'DASHSCOPE_API_KEY']  # muat API-key dari variabel lingkungan DASHSCOPE_API_KEY
    else:
        dashscope.api_key = '<your-dashscope-api-key>'  # atur API-key secara manual


# Callback pengenalan ucapan real-time
class Callback(RecognitionCallback):
    def on_open(self) -> None:
        global mic
        global stream
        print('RecognitionCallback open.')
        mic = pyaudio.PyAudio()
        stream = mic.open(format=pyaudio.paInt16,
                          channels=1,
                          rate=16000,
                          input=True)

    def on_close(self) -> None:
        global mic
        global stream
        print('RecognitionCallback close.')
        stream.stop_stream()
        stream.close()
        mic.terminate()
        stream = None
        mic = None

    def on_complete(self) -> None:
        print('RecognitionCallback completed.')  # pengenalan selesai

    def on_error(self, message) -> None:
        print('RecognitionCallback task_id: ', message.request_id)
        print('RecognitionCallback error: ', message.message)
        # Hentikan dan tutup aliran audio jika sedang berjalan
        if 'stream' in globals() and stream.active:
            stream.stop()
            stream.close()
        # Keluar paksa dari program
        sys.exit(1)

    def on_event(self, result: RecognitionResult) -> None:
        sentence = result.get_sentence()
        if 'text' in sentence:
            print('RecognitionCallback text: ', sentence['text'])
            if RecognitionResult.is_sentence_end(sentence):
                print(
                    'RecognitionCallback akhir kalimat, request_id:%s, usage:%s'
                    % (result.get_request_id(), result.get_usage(sentence)))


def signal_handler(sig, frame):
    print('Ctrl+C ditekan, hentikan pengenalan ...')
    # Hentikan pengenalan
    recognition.stop()
    print('Pengenalan dihentikan.')
    print(
        '[Metric] requestId: {}, delay paket pertama ms: {}, delay paket terakhir ms: {}'
        .format(
            recognition.get_last_request_id(),
            recognition.get_first_package_delay(),
            recognition.get_last_package_delay(),
        ))
    # Keluar paksa dari program
    sys.exit(0)


# fungsi utama
if __name__ == '__main__':
    init_dashscope_api_key()
    print('Menginisialisasi ...')

    # Buat callback pengenalan
    callback = Callback()

    # Panggil layanan pengenalan secara mode async, Anda dapat menyesuaikan parameter pengenalan, seperti model, format,
    # sample_rate
    recognition = Recognition(
        model='paraformer-realtime-v2',
        format=format_pcm,
        # 'pcm', 'wav', 'opus', 'speex', 'aac', atau 'amr'. Anda dapat memeriksa format yang didukung di dokumen.
        sample_rate=sample_rate,
        # 8000 atau 16000 didukung.
        semantic_punctuation_enabled=False,
        callback=callback)

    # Mulai pengenalan
    recognition.start()

    signal.signal(signal.SIGINT, signal_handler)
    print("Tekan 'Ctrl+C' untuk menghentikan perekaman dan pengenalan...")
    # Buat pendengar keyboard hingga "Ctrl+C" ditekan

    while True:
        if stream:
            data = stream.read(3200, exception_on_overflow=False)
            recognition.send_audio_frame(data)
        else:
            break

    recognition.stop()

Mengenali file audio lokal

File audio yang digunakan dalam contoh adalah: asr_example.wav.

import os
import time
from dashscope.audio.asr import *

# Jika Anda belum mengonfigurasi API key di variabel lingkungan, hapus komentar baris kode berikut dan ganti apiKey dengan API key Anda.
# import dashscope
# dashscope.api_key = "apiKey"

from datetime import datetime

def get_timestamp():
    now = datetime.now()
    formatted_timestamp = now.strftime("[%Y-%m-%d %H:%M:%S.%f]")
    return formatted_timestamp

class Callback(RecognitionCallback):
    def on_complete(self) -> None:
        print(get_timestamp() + ' Pengenalan selesai')  # pengenalan selesai

    def on_error(self, result: RecognitionResult) -> None:
        print('Tugas pengenalan: ', result.request_id)
        print('Error pengenalan: ', result.message)
        exit(0)

    def on_event(self, result: RecognitionResult) -> None:
        sentence = result.get_sentence()
        if 'text' in sentence:
            print(get_timestamp() + ' RecognitionCallback text: ', sentence['text'])
            if RecognitionResult.is_sentence_end(sentence):
                print(get_timestamp() + 
                    'RecognitionCallback akhir kalimat, request_id:%s, usage:%s'
                    % (result.get_request_id(), result.get_usage(sentence)))


callback = Callback()

recognition = Recognition(model='paraformer-realtime-v2',
                          format='wav',
                          sample_rate=16000,
                          # Parameter "language_hints" hanya didukung oleh model paraformer-realtime-v2.
                          language_hints=['zh', 'en'],
                          callback=callback)

recognition.start()

try:
    audio_data: bytes = None
    f = open("asr_example.wav", 'rb')
    if os.path.getsize("asr_example.wav"):
        while True:
            audio_data = f.read(3200)
            if not audio_data:
                break
            else:
                recognition.send_audio_frame(audio_data)
            time.sleep(0.1)
    else:
        raise Exception(
            'File yang disediakan kosong (berukuran nol byte)')
    f.close()
except Exception as e:
    raise e

recognition.stop()

print(
    '[Metric] requestId: {}, delay paket pertama ms: {}, delay paket terakhir ms: {}'
    .format(
        recognition.get_last_request_id(),
        recognition.get_first_package_delay(),
        recognition.get_last_package_delay(),
    ))