BatchalignHK 0.7.19.post6__tar.gz → 0.7.19.post7__tar.gz

{batchalignhk-0.7.19.post6 → batchalignhk-0.7.19.post7}/BatchalignHK.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.2
+Metadata-Version: 2.4
 Name: BatchalignHK
-Version: 0.7.19.post6
+Version: 0.7.19.post7
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu
@@ -50,6 +50,7 @@ Dynamic: author-email
 Dynamic: classifier
 Dynamic: description
 Dynamic: description-content-type
+Dynamic: license-file
 Dynamic: provides-extra
 Dynamic: requires-dist
 Dynamic: summary

{batchalignhk-0.7.19.post6 → batchalignhk-0.7.19.post7}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.2
+Metadata-Version: 2.4
 Name: BatchalignHK
-Version: 0.7.19.post6
+Version: 0.7.19.post7
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu
@@ -50,6 +50,7 @@ Dynamic: author-email
 Dynamic: classifier
 Dynamic: description
 Dynamic: description-content-type
+Dynamic: license-file
 Dynamic: provides-extra
 Dynamic: requires-dist
 Dynamic: summary

{batchalignhk-0.7.19.post6 → batchalignhk-0.7.19.post7}/batchalign/models/utterance/cantonese_infer.py

@@ -59,7 +59,7 @@ class BertCantoneseUtteranceModel(object):
         
         
         # Step 2: Define keywords and split the passage based on them
-        keywords = ['呀', '啦', '喎', '嘞', '㗎喇', '囉', '㗎', '啊', '嗯']  # Replace with your desired keywords
+        keywords = ['呀', '喎', '嘞', '㗎喇', '囉', '啊', '嗯']  # Replace with your desired keywords
         
         chunks = []
         start = 0

batchalignhk-0.7.19.post7/batchalign/pipelines/asr/tencent.py

@@ -0,0 +1,246 @@
+"""
+rev.py
+Support for Rev.ai, a commerical ASR service
+"""
+
+from batchalign.document import *
+from batchalign.pipelines.base import *
+from batchalign.pipelines.asr.utils import *
+from batchalign.utils.config import config_read
+
+from batchalign.errors import *
+
+from batchalign.models import BertUtteranceModel, BertCantoneseUtteranceModel, resolve
+
+from opencc import OpenCC
+cc = OpenCC('s2hk')
+
+import time
+import pathlib
+import tempfile
+import pycountry
+import numpy as np
+import soundfile as sf
+from pydub import AudioSegment
+from pydub.effects import normalize
+import base64
+from tencentcloud.common.credential import Credential
+from tencentcloud.asr.v20190614.asr_client import AsrClient, models
+
+import asyncio
+import tempfile
+import os
+from pydub import AudioSegment
+from pydub.effects import normalize
+from pydub.exceptions import CouldntDecodeError
+
+
+import logging
+L = logging.getLogger("batchalign")
+
+class TencentEngine(BatchalignEngine):
+
+    @property
+    def tasks(self):
+        if self.__engine:
+            return [ Task.ASR, Task.SPEAKER_RECOGNITION, Task.UTTERANCE_SEGMENTATION ]
+        else:
+            return [ Task.ASR, Task.SPEAKER_RECOGNITION ]
+
+    def __init__(self, key:str=None, lang="eng", num_speakers=2):
+
+        if key == None or key.strip() == "":
+            config = config_read()
+            try:
+                id = config["asr"]["engine.tencent.id"] 
+                key = config["asr"]["engine.tencent.key"] 
+            except KeyError:
+                raise ConfigError("No Tencent Cloud key found. Tencent Cloud was not set up! Please write one yourself and place it at ~/.batchalign.ini.")
+
+        self.__lang_code = lang
+        self.__num_speakers = num_speakers
+
+        if lang == "yue":
+            self.__lang = "yue"
+        else:
+            self.__lang = pycountry.languages.get(alpha_3=lang).alpha_2
+
+        cred = Credential(id, key)
+        self.__client = AsrClient(cred, "ap-hongkong")
+
+        if resolve("utterance", lang) != None:
+            L.debug("Initializing utterance model...")
+            if lang != "yue":
+                self.__engine = BertUtteranceModel(resolve("utterance", lang))
+            else:
+                self.__engine = BertCantoneseUtteranceModel(resolve("utterance", lang))
+            L.debug("Done.")
+        else:
+            self.__engine = None
+
+    def __preprocess_audio(self, input_path):
+        """Enhanced audio preprocessing for low-volume speech"""
+        try:
+            L.info(f"Optimizing audio for ASR: {input_path}")
+            
+            # read the audio file
+            audio = AudioSegment.from_file(input_path)
+
+            audio = audio.set_channels(1)  
+            audio = audio.set_frame_rate(16000)  
+
+            
+            audio = audio.compress_dynamic_range(
+                threshold=-40, 
+                ratio=3,
+                attack=5, 
+                release=100
+            )
+            audio = audio.low_pass_filter(4000)  # filter out high frequencies
+            audio = audio.normalize(headroom=2)  # keep the headroom
+            audio = audio.compress_dynamic_range(
+                threshold=-55,
+                ratio=6,
+                attack=15,
+                release=200
+            )
+
+            # enhance low volume
+            audio = audio.high_pass_filter(80)
+            boosted = audio.high_pass_filter(1000).apply_gain(+4)
+            audio = audio.overlay(boosted)
+
+            if L.level <= logging.DEBUG:
+                self.__print_audio_stats(audio)
+
+            # output to a temporary file
+            temp_fd, temp_path = tempfile.mkstemp(suffix=".mp3")
+            os.close(temp_fd)
+            audio.export(
+                temp_path,
+                format="mp3",
+                codec="libmp3lame",
+                bitrate="96k",
+                tags={"title": "BA_Optimized"},
+                parameters=[
+                    "-compression_level", "2",
+                    "-reservoir", "0",
+                    "-joint_stereo", "0"
+                ]
+            )
+
+            return temp_path
+
+        except CouldntDecodeError:
+            L.error(f"Audio decoding failed: {input_path}")
+            return input_path
+        except Exception as e:
+            L.error(f"Audio processing error: {str(e)}")
+            return input_path
+        
+    def replace_cantonese_words(self, word):
+        """Function to replace Cantonese words with custom replacements."""
+        word_replacements = {
+            "系": "係", 
+            "唔系": "唔係",
+            "噶": "㗎",
+            "咧": "呢",
+            "嗬": "喎",
+            "只": "隻",
+            "咯": "囉",
+            "嚇": "吓",
+            "飲": "飲",
+            "喐": "郁",
+            "食": "食",
+            "啫": "咋",
+            "哇": "嘩",
+            "着": "著",
+            "中意": "鍾意",
+            "嘞": "喇",
+            "啵": "噃",
+            "遊水": "游水",
+            "羣組": "群組",
+            "古仔": "故仔",
+            "甕": "㧬",
+            "牀": "床",
+            "松": "鬆",
+            "較剪": "鉸剪",
+            "吵": "嘈",
+            "衝涼": "沖涼",
+            "分鍾": "分鐘",
+            "重復": "重複"
+        }
+        return word_replacements.get(word, word)
+    
+    def generate(self, f, **kwargs):
+        lang = self.__lang
+        client = self.__client
+
+        processed_path = self.__preprocess_audio(f)
+        audio = AudioSegment.from_file(processed_path)
+        
+        try:
+            L.info(f"Uploading '{pathlib.Path(f).stem}'...")
+            with open(processed_path, "rb") as audio_file:
+                encoded_string = base64.b64encode(audio_file.read())
+
+            req = models.CreateRecTaskRequest()
+            if lang in {'zho', 'yue', 'wuu', 'nan','hak'}:
+                req.EngineModelType = "16k_zh_large"
+            else:
+                req.EngineModelType = f"16k_{lang}"
+            req.ResTextFormat = 1
+            req.SpeakerDiarization = 1
+            req.ChannelNum = 1
+            req.Data = encoded_string.decode('ascii')
+            req.SourceType = 1
+            
+            resp = client.CreateRecTask(req)
+
+            L.info(f"Tencent is transcribing '{pathlib.Path(f).stem}'...")
+            req = models.DescribeTaskStatusRequest()
+            req.TaskId = resp.Data.TaskId
+
+            res = client.DescribeTaskStatus(req)
+            while res.Data.Status not in [2, 3]:
+                time.sleep(15)
+                res = client.DescribeTaskStatus(req)
+
+            if res.Data.Status in ["3", 3]:
+                raise RuntimeError(f"Tencent reports job failed! error='{res.Data.ErrorMsg}'")
+
+            turns = []
+            for i in res.Data.ResultDetail:
+                turn = []
+                start = i.StartMs
+                for j in i.Words:
+                    word = j.Word
+                    if self.__lang == "yue":
+                        word = cc.convert(word)
+                        
+                        word = self.replace_cantonese_words(word)
+
+                    turn.append({
+                        "type": "text",
+                        "ts": (j.OffsetStartMs + start) / 1000,
+                        "end_ts": (j.OffsetEndMs + start) / 1000,
+                        "value": word
+                    })
+                turns.append({
+                    "elements": turn,
+                    "speaker": i.SpeakerId
+                })
+            L.debug(f"Tencent done.")
+            
+            # Extract the text from the small volume parts for translation
+
+            doc = process_generation({"monologues": turns},
+                                    self.__lang_code, 
+                                    utterance_engine=self.__engine)
+            media = Media(type=MediaType.AUDIO, name=Path(f).stem, url=f)
+            doc.media = media
+            return doc
+            
+        finally:
+            if processed_path != f and pathlib.Path(processed_path).exists():
+                pathlib.Path(processed_path).unlink()

{batchalignhk-0.7.19.post6 → batchalignhk-0.7.19.post7}/batchalign/version

@@ -1,3 +1,3 @@
-0.7.19-post.6
+0.7.19-post.7
 May 20th, 2025
 fixes for ASR

batchalignhk-0.7.19.post6/batchalign/pipelines/asr/tencent.py

@@ -1,132 +0,0 @@
-"""
-rev.py
-Support for Rev.ai, a commerical ASR service
-"""
-
-from batchalign.document import *
-from batchalign.pipelines.base import *
-from batchalign.pipelines.asr.utils import *
-from batchalign.utils.config import config_read
-
-from batchalign.errors import *
-
-from batchalign.models import BertUtteranceModel, BertCantoneseUtteranceModel, resolve
-
-from opencc import OpenCC
-cc = OpenCC('s2hk')
-
-import time
-import pathlib
-import pycountry
-
-import base64
-from tencentcloud.common.credential import Credential
-from tencentcloud.asr.v20190614.asr_client import AsrClient, models
-
-import logging
-L = logging.getLogger("batchalign")
-
-class TencentEngine(BatchalignEngine):
-
-    @property
-    def tasks(self):
-        # if there is no utterance segmentation scheme, we only
-        # run ASR
-        if self.__engine:
-            return [ Task.ASR, Task.SPEAKER_RECOGNITION, Task.UTTERANCE_SEGMENTATION ]
-        else:
-            return [ Task.ASR, Task.SPEAKER_RECOGNITION ]
-
-    def __init__(self, key:str=None, lang="eng", num_speakers=2):
-
-        if key == None or key.strip() == "":
-            config = config_read()
-            try:
-                id = config["asr"]["engine.tencent.id"] 
-                key = config["asr"]["engine.tencent.key"] 
-            except KeyError:
-                raise ConfigError("No Tencent Cloud key found. Tencent Cloud was not set up! Please write one yourself and place it at `~/.batchalign.ini`.")
-
-        self.__lang_code = lang
-        self.__num_speakers = num_speakers
-
-        if lang == "yue":
-            self.__lang = "yue"
-        else:
-            self.__lang = pycountry.languages.get(alpha_3=lang).alpha_2
-
-        cred = Credential(id,key)
-        self.__client = AsrClient(cred, "ap-hongkong")
-
-        if resolve("utterance", lang) != None:
-            L.debug("Initializing utterance model...")
-            if lang != "yue":
-                self.__engine = BertUtteranceModel(resolve("utterance", lang))
-            else:
-                # we have special inference procedure for cantonese
-                self.__engine = BertCantoneseUtteranceModel(resolve("utterance", lang))
-            L.debug("Done.")
-        else:
-            self.__engine = None
-
-
-    def generate(self, f, **kwargs):
-        # bring language code into the stack to access
-        lang = self.__lang
-        client = self.__client
-
-        L.info(f"Uploading '{pathlib.Path(f).stem}'...")
-        # we will send the file for processing
-        if not str(f).startswith("http"):
-            with open(f, "rb") as image_file:
-                encoded_string = base64.b64encode(image_file.read())
-
-        req = models.CreateRecTaskRequest()
-        req.EngineModelType = f"16k_{lang}"
-        req.ResTextFormat = 1
-        req.SpeakerDiarization = 1
-        req.ChannelNum = 1
-        if not str(f).startswith("http"):
-            req.Data = encoded_string.decode('ascii')
-            req.SourceType = 1
-        else:
-            req.Url = f
-            req.SourceType = 0
-        resp = client.CreateRecTask(req)
-
-        L.info(f"Tencent is transcribing '{pathlib.Path(f).stem}'...")
-        req = models.DescribeTaskStatusRequest()
-        req.TaskId = resp.Data.TaskId
-
-        res = client.DescribeTaskStatus(req)
-        while res.Data.Status not in [2,3]:
-            time.sleep(15)
-            res = client.DescribeTaskStatus(req)
-
-        # if failed, raise
-        if res.Data.Status == "3" or res.Data.Status == 3:
-            raise RuntimeError(f"Tencent reports job failed! error='{res.Data.ErrorMsg}'")
-
-        turns = []
-        for i in res.Data.ResultDetail:
-            turn = []
-            start = i.StartMs
-            for j in i.Words:
-                turn.append({
-                    "type": "text",
-                    "ts": (j.OffsetStartMs+start)/1000,
-                    "end_ts": (j.OffsetEndMs+start)/1000,
-                    "value": cc.convert(j.Word) if self.__lang == "yue" else j.Word
-                })
-            turns.append({
-                "elements": turn,
-                "speaker": i.SpeakerId
-            })
-        L.debug(f"Tencent done.")
-
-        # postprocess the output and define media tier
-        doc = process_generation({"monologues": turns},
-                                 self.__lang_code, utterance_engine=self.__engine)
-        media = Media(type=MediaType.AUDIO, name=Path(f).stem, url=f)
-        doc.media = media
-        return doc