[+] Telegram bot

.gitignore

@@ -117,3 +117,6 @@ dmypy.json
 # Custom
 .idea
 ._*
+voice-bot-token.txt
+audio_tmp
+src/voice-bot-token.txt

requirements.txt

@@ -2,3 +2,5 @@
 tensorflow==2.2.0
 plaidml-keras==0.7.0
 inaSpeechSegmenter==0.6.8
+
+python-telegram-bot

src/bot.py

@@ -0,0 +1,113 @@
+import os
+from datetime import datetime
+from pathlib import Path
+from telegram import Update, Message
+from telegram.ext import Updater, CallbackContext, Dispatcher, CommandHandler, MessageHandler, \
+    Filters
+
+os.environ['KERAS_BACKEND'] = "plaidml.keras.backend"
+
+import keras
+
+from inaSpeechSegmenter import *
+from ina_main import *
+import warnings
+
+warnings.filterwarnings("ignore")
+
+
+def r(u: Update, msg: str, md=True):
+    updater.bot.sendMessage(chat_id=u.effective_chat.id, text=msg,
+                            parse_mode='Markdown' if md else None)
+
+
+def cmd_start(u: Update, c: CallbackContext):
+    r(u, '欢迎! 点下面的录音按钮就可以开始啦w')
+
+
+def process_audio(message: Message):
+    # Only when replying to voice or audio
+    audio = message.audio or message.voice
+    if not audio:
+        return
+
+    # Download audio file
+    date = datetime.now().strftime('%Y-%m-%d %H-%M')
+    downloader = bot.getFile(audio.file_id)
+    file = Path(tmpdir).joinpath(f'{date} {message.from_user.name[1:]}.mp3')
+    print(downloader, '->', file)
+    downloader.download(file)
+
+    # Segment file
+    seg = Segmenter()
+    result = process(seg, [str(file.absolute())]).results[0]
+
+    # Null case
+    print(result.frames)
+    if len(result.frames) == 0:
+        bot.send_message(message.chat_id, '分析失败, 大概是音量太小或者时长太短吧, 再试试w')
+        return
+
+    # Draw results
+    with draw_result(str(file), result) as buf:
+        f, m, o = get_result_percentages(result)
+        msg = f"分析结果: {f*100:.0f}% 🙋‍♀️ | {m*100:.0f}% 🙋‍♂️ | {o*100:.0f}% 🚫"
+        bot.send_photo(message.chat_id, photo=buf, caption=msg,
+                       reply_to_message_id=message.message_id)
+
+
+def cmd_analyze(u: Update, c: CallbackContext):
+    reply = u.effective_message.reply_to_message
+
+    # Parse command
+    text = u.effective_message.text
+    if not text:
+        return
+    cmd = text.lower().split()[0].strip()
+
+    if cmd[0] not in '!/':
+        return
+    cmd = cmd[1:]
+
+    if cmd not in ['analyze', 'analyze-raw']:
+        return
+
+    if cmd == 'analyze-raw':
+        raw = True
+
+    if u.effective_user.id == reply.from_user.id:
+        process_audio(reply)
+    else:
+        r(u, '只有自己能分析自己的音频哦 👀')
+
+
+def on_audio(u: Update, c: CallbackContext):
+    process_audio(u.effective_message)
+
+
+if __name__ == '__main__':
+    tmpdir = Path('audio_tmp')
+    tmpdir.mkdir(exist_ok=True, parents=True)
+
+    # Find telegram token
+    path = Path(os.path.abspath(__file__)).parent
+    db_path = path.joinpath('voice-bot-db.json')
+    if 'tg_token' in os.environ:
+        tg_token = os.environ['tg_token']
+    else:
+        with open(path.joinpath('voice-bot-token.txt'), 'r', encoding='utf-8') as f:
+            tg_token = f.read().strip()
+
+    # Telegram login
+    updater = Updater(token=tg_token, use_context=True)
+    dispatcher: Dispatcher = updater.dispatcher
+    bot = updater.bot
+
+    dispatcher.add_handler(CommandHandler('start', cmd_start, filters=Filters.chat_type.private))
+    dispatcher.add_handler(CommandHandler('analyze', cmd_analyze, filters=Filters.reply))
+    dispatcher.add_handler(MessageHandler(Filters.reply, cmd_analyze))
+    dispatcher.add_handler(MessageHandler(Filters.voice & Filters.chat_type.private, on_audio))
+    dispatcher.add_handler(MessageHandler(Filters.audio & Filters.chat_type.private, on_audio))
+
+    print('Starting bot...')
+    updater.start_polling()

src/ina_main.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import io
 import os
+import shutil
 import sys
 import tempfile
 import time
@@ -62,7 +63,7 @@ def process(self: Segmenter, inp: list[str], tmpdir=None, verbose=False, skip_if
             break
         mspec, loge, diff_len = feats
         lseg = self.segment_feats(mspec, loge, diff_len, 0)
-        results.append(Result(lseg, inp[len(lmsg) - 1]))
+        results.append(Result([ResultFrame(*s) for s in lseg], inp[len(lmsg) - 1]))
 
     t_batch_dur = time.time() - t_batch_start
     nb_processed = len([e for e in lmsg if e[1] == 0])
@@ -93,6 +94,8 @@ def to_wav(file: str, callback: Callable, start_sec: float = 0, stop_sec: float
         output, error = p.communicate()
         assert p.returncode == 0, error
 
+        shutil.copy2(tmp_wav, './test_leohearts.wav')
+
         return callback(tmp_wav)
 
 
@@ -130,7 +133,7 @@ def draw_result(file: str, result: Result):
         ax.set_xlim([result.frames[0].start, result.frames[-1].end])
 
         # Draw segmentation areas
-        colors = {'female': '#F5A9B8', 'male': '#5BCEFA', 'default': 'white'}
+        colors = {'female': '#F5A9B8', 'male': '#5BCEFA', 'default': 'gray'}
         for r in result.frames:
             color = colors[r.gender] if r.gender in colors else colors['default']
             ax.axvspan(r.start, r.end - 0.01, alpha=.5, color=color)
@@ -140,6 +143,8 @@ def draw_result(file: str, result: Result):
         plt.axis('off')
         plt.savefig(buf, bbox_inches='tight', pad_inches=0, transparent=False)
         buf.seek(0)
+        plt.clf()
+        plt.close()
         return buf
 
     return to_wav(file, wav_callback)
@@ -165,27 +170,26 @@ def get_result_percentages(result: Result) -> tuple[float, float, float]:
         durations[d] /= total_dur
 
     # Return results
-    f = durations['female']
-    m = durations['male']
+    f = durations.get('female', 0)
+    m = durations.get('male', 0)
     return f, m, 1 - f - m
 
 
+# def test():
+#     results: BatchResults = BatchResults(
+#         [Result([ResultFrame('female', 0.0, 10.48), ResultFrame('male', 10.48, 12.780000000000001)],
+#                 '../test.csv')],
+#         1.7032792568206787, 1.7032792568206787, 1,
+#         [('../test.csv', 0)])
+#
+#     with draw_result('../test.mp3', results.results[0]) as buf:
+#         show_image_buffer(buf)
+#         print(get_result_percentages(results.results[0]))
+#
+#     # seg = Segmenter()
+#     # print(process(seg, ['../test.mp3']))
+
 if __name__ == '__main__':
-    args = sys.argv[1:]
-    if len(args) < 0:
-        exit(0)
-
-    results: BatchResults = BatchResults(
-        [Result([ResultFrame('female', 0.0, 10.48), ResultFrame('male', 10.48, 12.780000000000001)],
-                '../test.csv')],
-        1.7032792568206787, 1.7032792568206787, 1,
-        [('../test.csv', 0)])
-
-    with draw_result('../test.mp3', results.results[0]) as buf:
-        show_image_buffer(buf)
-        print(get_result_percentages(results.results[0]))
-
-    # inp = args[0]
-
-    # seg = Segmenter()
-    # print(process(seg, ['../test.mp3']))
+    to_wav('../audio_tmp/2021-12-22 05-32 leph1art5.mp3', print)
+    # test()
+    pass