|
| 1 | +# coding=utf-8 |
| 2 | + |
| 3 | +""" |
| 4 | +requires Python 3.6 or later |
| 5 | +
|
| 6 | +pip install asyncio |
| 7 | +pip install websockets |
| 8 | +""" |
| 9 | + |
| 10 | +import base64 |
| 11 | +import json |
| 12 | +import os |
| 13 | +import time |
| 14 | +import uuid |
| 15 | +import requests |
| 16 | +import uuid_utils.compat as uuid |
| 17 | + |
| 18 | +from typing import Dict |
| 19 | + |
| 20 | +from common.utils.logger import maxkb_logger |
| 21 | +from models_provider.base_model_provider import MaxKBBaseModel |
| 22 | +from models_provider.impl.base_stt import BaseSpeechToText |
| 23 | + |
| 24 | +audio_format = "mp3" # wav 或者 mp3，根据实际音频格式设置 |
| 25 | + |
| 26 | + |
| 27 | +def determine_api_mode(url): |
| 28 | + """ |
| 29 | + 根据URL判断API模式 |
| 30 | + """ |
| 31 | + if '/recognize/flash' in url: |
| 32 | + return 'sync' |
| 33 | + elif '/submit' in url: |
| 34 | + return 'async_submit' |
| 35 | + elif '/query' in url: |
| 36 | + return 'async_query' |
| 37 | + else: |
| 38 | + return 'unknown' |
| 39 | + |
| 40 | + |
| 41 | +class VolcanicASRClient: |
| 42 | + def __init__(self, appid, token): |
| 43 | + self.appid = appid |
| 44 | + self.token = token |
| 45 | + |
| 46 | + def _build_headers(self, url, task_id=None, x_tt_logid=None): |
| 47 | + """根据URL构建请求头""" |
| 48 | + mode = determine_api_mode(url) |
| 49 | + |
| 50 | + headers = { |
| 51 | + "X-Api-App-Key": self.appid, |
| 52 | + "X-Api-Access-Key": self.token, |
| 53 | + } |
| 54 | + |
| 55 | + if mode == 'sync': |
| 56 | + headers.update({ |
| 57 | + "X-Api-Resource-Id": "volc.bigasr.auc_turbo", |
| 58 | + "X-Api-Request-Id": str(uuid.uuid4()), |
| 59 | + "X-Api-Sequence": "-1", |
| 60 | + }) |
| 61 | + elif mode == 'async_submit': |
| 62 | + headers.update({ |
| 63 | + "X-Api-Resource-Id": "volc.bigasr.auc", |
| 64 | + "X-Api-Request-Id": task_id or str(uuid.uuid4()), |
| 65 | + "X-Api-Sequence": "-1", |
| 66 | + }) |
| 67 | + elif mode == 'async_query': |
| 68 | + headers.update({ |
| 69 | + "X-Api-Resource-Id": "volc.bigasr.auc", |
| 70 | + "X-Api-Request-Id": task_id or str(uuid.uuid4()), |
| 71 | + "X-Tt-Logid": x_tt_logid or "", |
| 72 | + }) |
| 73 | + |
| 74 | + return headers |
| 75 | + |
| 76 | + def _create_request_body(self, audio_data, mode='sync'): |
| 77 | + """创建请求体""" |
| 78 | + base_request = { |
| 79 | + "user": {"uid": self.appid if mode == 'sync' else "fake_uid"}, |
| 80 | + "audio": audio_data, |
| 81 | + } |
| 82 | + |
| 83 | + if mode == 'sync': |
| 84 | + base_request["request"] = { |
| 85 | + "model_name": "bigmodel", |
| 86 | + "enable_itn": True, |
| 87 | + "enable_punc": True, |
| 88 | + "enable_ddc": True, |
| 89 | + } |
| 90 | + else: # async |
| 91 | + base_request["request"] = { |
| 92 | + "model_name": "bigmodel", |
| 93 | + "enable_channel_split": True, |
| 94 | + "enable_ddc": True, |
| 95 | + "enable_speaker_info": True, |
| 96 | + "enable_punc": True, |
| 97 | + "enable_itn": True, |
| 98 | + "corpus": { |
| 99 | + "correct_table_name": "", |
| 100 | + "context": "" |
| 101 | + } |
| 102 | + } |
| 103 | + |
| 104 | + return base_request |
| 105 | + |
| 106 | + def process_audio(self, audio_file=None, submit_url=None): |
| 107 | + """ |
| 108 | + 根据submit_url自动选择处理模式 |
| 109 | + """ |
| 110 | + # 获取音频数据 |
| 111 | + base64_audio = base64.b64encode(audio_file.read()).decode("utf-8") |
| 112 | + audio_data = {"data": base64_audio} |
| 113 | + |
| 114 | + # 根据URL判断API模式 |
| 115 | + mode = determine_api_mode(submit_url) |
| 116 | + |
| 117 | + if mode == 'sync': |
| 118 | + return self._sync_recognize(audio_data, submit_url) |
| 119 | + elif mode == 'async_submit': |
| 120 | + return self._async_process(audio_data, submit_url) |
| 121 | + else: |
| 122 | + raise ValueError(f"Unsupported URL pattern: {submit_url}") |
| 123 | + |
| 124 | + def _get_audio_data(self, audio_file): |
| 125 | + """构建音频数据对象""" |
| 126 | + base64_audio = base64.b64encode(audio_file.read()).decode("utf-8") |
| 127 | + return {"data": base64_audio} |
| 128 | + |
| 129 | + def _sync_recognize(self, audio_data, submit_url): |
| 130 | + """同步识别模式""" |
| 131 | + headers = self._build_headers(submit_url) |
| 132 | + request_body = self._create_request_body(audio_data, mode='sync') |
| 133 | + |
| 134 | + response = requests.post(submit_url, json=request_body, headers=headers) |
| 135 | + return self._handle_response(response, "sync_recognize") |
| 136 | + |
| 137 | + def _async_process(self, audio_data, submit_url): |
| 138 | + """异步处理模式""" |
| 139 | + # 提交任务 |
| 140 | + task_id = str(uuid.uuid4()) |
| 141 | + headers = self._build_headers(submit_url, task_id=task_id) |
| 142 | + request_body = self._create_request_body(audio_data, mode='async') |
| 143 | + |
| 144 | + submit_response = requests.post(submit_url, data=json.dumps(request_body), headers=headers) |
| 145 | + |
| 146 | + if submit_response.headers.get("X-Api-Status-Code") == "20000000": |
| 147 | + x_tt_logid = submit_response.headers.get("X-Tt-Logid", "") |
| 148 | + # 查询结果 |
| 149 | + return self._poll_for_result(task_id, x_tt_logid) |
| 150 | + else: |
| 151 | + print(f"Submit task failed: {submit_response.headers}") |
| 152 | + return None |
| 153 | + |
| 154 | + def _poll_for_result(self, task_id, x_tt_logid): |
| 155 | + """轮询查询异步任务结果""" |
| 156 | + query_url = "https://openspeech-direct.zijieapi.com/api/v3/auc/bigmodel/query" |
| 157 | + |
| 158 | + while True: |
| 159 | + query_response = self._query_task(task_id, x_tt_logid, query_url) |
| 160 | + code = query_response.headers.get('X-Api-Status-Code', "") |
| 161 | + |
| 162 | + if code == '20000000': # 任务完成 |
| 163 | + return query_response |
| 164 | + elif code != '20000001' and code != '20000002': # 任务失败 |
| 165 | + print(f"Async task failed with code: {code}") |
| 166 | + return None |
| 167 | + time.sleep(1) |
| 168 | + |
| 169 | + def _query_task(self, task_id, x_tt_logid, query_url): |
| 170 | + """执行单次查询请求""" |
| 171 | + headers = self._build_headers(query_url, task_id=task_id, x_tt_logid=x_tt_logid) |
| 172 | + response = requests.post(query_url, json.dumps({}), headers=headers) |
| 173 | + return self._handle_response(response, "async_query", silent=True) |
| 174 | + |
| 175 | + def _handle_response(self, response, operation, silent=False): |
| 176 | + """处理响应""" |
| 177 | + if 'X-Api-Status-Code' in response.headers: |
| 178 | + if not silent: |
| 179 | + print(f'{operation} response header X-Api-Status-Code: {response.headers["X-Api-Status-Code"]}') |
| 180 | + print(f'{operation} response header X-Api-Message: {response.headers["X-Api-Message"]}') |
| 181 | + print(f'{operation} response header X-Tt-Logid: {response.headers["X-Tt-Logid"]}') |
| 182 | + |
| 183 | + if operation == "sync_recognize": |
| 184 | + print(f'sync response content: {response.json()}\n') |
| 185 | + |
| 186 | + return response |
| 187 | + else: |
| 188 | + print(f'{operation} failed: {response.headers}\n') |
| 189 | + return None |
| 190 | + |
| 191 | + |
| 192 | +class VolcanicEngineBigModelSpeechToText(MaxKBBaseModel, BaseSpeechToText): |
| 193 | + volcanic_app_id: str |
| 194 | + volcanic_api_url: str |
| 195 | + volcanic_token: str |
| 196 | + params: dict |
| 197 | + |
| 198 | + def __init__(self, **kwargs): |
| 199 | + super().__init__(**kwargs) |
| 200 | + self.volcanic_api_url = kwargs.get('volcanic_api_url') |
| 201 | + self.volcanic_token = kwargs.get('volcanic_token') |
| 202 | + self.volcanic_app_id = kwargs.get('volcanic_app_id') |
| 203 | + self.params = kwargs.get('params') |
| 204 | + |
| 205 | + @staticmethod |
| 206 | + def is_cache_model(): |
| 207 | + return False |
| 208 | + |
| 209 | + @staticmethod |
| 210 | + def new_instance(model_type, model_name, model_credential: Dict[str, object], **model_kwargs): |
| 211 | + optional_params = {} |
| 212 | + if 'max_tokens' in model_kwargs and model_kwargs['max_tokens'] is not None: |
| 213 | + optional_params['max_tokens'] = model_kwargs['max_tokens'] |
| 214 | + if 'temperature' in model_kwargs and model_kwargs['temperature'] is not None: |
| 215 | + optional_params['temperature'] = model_kwargs['temperature'] |
| 216 | + return VolcanicEngineBigModelSpeechToText( |
| 217 | + volcanic_api_url=model_credential.get('volcanic_api_url'), |
| 218 | + volcanic_token=model_credential.get('volcanic_token'), |
| 219 | + volcanic_app_id=model_credential.get('volcanic_app_id'), |
| 220 | + params=model_kwargs, |
| 221 | + ) |
| 222 | + |
| 223 | + def check_auth(self): |
| 224 | + cwd = os.path.dirname(os.path.abspath(__file__)) |
| 225 | + with open(f'{cwd}/iat_mp3_16k.mp3', 'rb') as audio_file: |
| 226 | + self.speech_to_text(audio_file) |
| 227 | + |
| 228 | + def speech_to_text(self, audio_file): |
| 229 | + try: |
| 230 | + client = VolcanicASRClient(self.volcanic_app_id, self.volcanic_token) |
| 231 | + result = client.process_audio(audio_file, self.volcanic_api_url) |
| 232 | + if result.status_code == 200: |
| 233 | + return result.json().get('result').get('text') |
| 234 | + except Exception as e: |
| 235 | + maxkb_logger.error(f'Error getting speech to text: {e}') |
0 commit comments