BatchalignHK 0.7.19.post13__tar.gz → 0.7.19.post15__tar.gz

{batchalignhk-0.7.19.post13 → batchalignhk-0.7.19.post15}/BatchalignHK.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.4
+Metadata-Version: 2.2
 Name: BatchalignHK
-Version: 0.7.19.post13
+Version: 0.7.19.post15
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu
@@ -50,7 +50,6 @@ 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.post13 → batchalignhk-0.7.19.post15}/BatchalignHK.egg-info/SOURCES.txt

@@ -92,6 +92,7 @@ batchalign/pipelines/translate/seamless.py
 batchalign/pipelines/translate/utils.py
 batchalign/pipelines/utr/__init__.py
 batchalign/pipelines/utr/rev_utr.py
+batchalign/pipelines/utr/tencent_utr.py
 batchalign/pipelines/utr/utils.py
 batchalign/pipelines/utr/whisper_utr.py
 batchalign/pipelines/utterance/__init__.py

{batchalignhk-0.7.19.post13 → batchalignhk-0.7.19.post15}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.4
+Metadata-Version: 2.2
 Name: BatchalignHK
-Version: 0.7.19.post13
+Version: 0.7.19.post15
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu
@@ -50,7 +50,6 @@ 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.post13 → batchalignhk-0.7.19.post15}/batchalign/cli/cli.py

@@ -110,10 +110,16 @@ batchalign.add_command(train, "models")
               default=False, help="For utterance timing recovery, OpenAI Whisper (ASR) instead of Rev.AI (default).")
 @click.option("--wav2vec/--whisper_fa",
               default=True, help="Use Whisper instead of Wav2Vec for English (defaults for Whisper for non-English)")
+@click.option("--tencent/--rev",
+              default=False, help="Use Tencent instead of Rev.AI (default).")
 @click.option("--pauses", type=bool, default=False, help="Should we try to bullet each word or should we try to add pauses in between words by grouping them? Default: no pauses.", is_flag=True)
-
+@click.option("--wor/--nowor",
+              default=True, help="Should we write word level alignment line? Default to yes.")
+@click.option("--data",
+              help="the URL of the data",
+              type=str)
 @click.pass_context
-def align(ctx, in_dir, out_dir, whisper, wav2vec, **kwargs):
+def align(ctx, in_dir, out_dir, whisper, wav2vec, tencent, **kwargs):
     """Align transcripts against corresponding media files."""
     def loader(file):
         return (
@@ -122,7 +128,7 @@ def align(ctx, in_dir, out_dir, whisper, wav2vec, **kwargs):
         )
 
     def writer(doc, output):
-        CHATFile(doc=doc).write(output)
+        CHATFile(doc=doc).write(output, write_wor=kwargs.get("wor", True))
 
     if not wav2vec:
         _dispatch("align", "eng", 1,
@@ -130,7 +136,8 @@ def align(ctx, in_dir, out_dir, whisper, wav2vec, **kwargs):
                   in_dir, out_dir,
                   loader, writer, C,
                   fa="whisper_fa",
-                  utr="whisper_utr" if whisper else "rev_utr",
+                  utr=("whisper_utr" if whisper else
+                       ("tencent_utr" if tencent else "rev_utr")),
                   **kwargs)
     else:
         _dispatch("align", "eng", 1,
@@ -138,7 +145,8 @@ def align(ctx, in_dir, out_dir, whisper, wav2vec, **kwargs):
                   in_dir, out_dir,
                   loader, writer, C,
                   fa="wav2vec_fa",
-                  utr="whisper_utr" if whisper else "rev_utr",
+                  utr=("whisper_utr" if whisper else
+                       ("tencent_utr" if tencent else "rev_utr")),
                   **kwargs)
 
 #################### TRANSCRIBE ################################

{batchalignhk-0.7.19.post13 → batchalignhk-0.7.19.post15}/batchalign/cli/dispatch.py

@@ -77,6 +77,8 @@ def _dispatch(command, lang, num_speakers,
             files.append(url)
             outputs.append(os.path.join(out_dir, base))
 
+    extr_data_mapping = {}
+
     for basedir, _, fs in os.walk(in_dir):
         for f in fs:
             path = Path(os.path.join(basedir, f))
@@ -116,6 +118,16 @@ def _dispatch(command, lang, num_speakers,
             # if the file needs to get processed, append it to the list
             # to be processed and compute the output 
             if ext in extensions:
+                for indx, i in enumerate(files):
+                    # check if this is a duplicate file
+                    if (not isinstance(i, str) and
+                        Path(i.geturl()).stem == Path(inp_path).stem):
+                        extr_data_mapping[inp_path] = i.geturl()
+
+                        files.pop(indx)
+                        outputs.pop(indx)
+                        break
+                        
                 files.append(inp_path)
                 outputs.append(str(repathed))
             # otherwise just copy the file
@@ -176,6 +188,7 @@ def _dispatch(command, lang, num_speakers,
                     # RUN THE PUPPY!
                     doc = pipeline(doc,
                                    callback=lambda *args:progress_callback(file, *args),
+                                   extra_info={"extra_input": extr_data_mapping.get(file)},
                                    **kw)
                 msgs = [escape(str(i.message)).strip() for i in w]
                 # write the format, as needed

{batchalignhk-0.7.19.post13 → batchalignhk-0.7.19.post15}/batchalign/pipelines/__init__.py

@@ -7,7 +7,7 @@ from .cleanup import NgramRetraceEngine, DisfluencyReplacementEngine
 from .speaker import NemoSpeakerEngine
 
 from .fa import WhisperFAEngine, Wave2VecFAEngine
-from .utr import WhisperUTREngine, RevUTREngine
+from .utr import WhisperUTREngine, RevUTREngine, TencentUTREngine
 
 from .analysis import EvaluationEngine
 from .utterance import StanzaUtteranceEngine

{batchalignhk-0.7.19.post13 → batchalignhk-0.7.19.post15}/batchalign/pipelines/dispatch.py

@@ -7,7 +7,7 @@ from batchalign import (WhisperEngine, WhisperFAEngine, StanzaEngine, RevEngine,
                         NgramRetraceEngine, DisfluencyReplacementEngine, WhisperUTREngine,
                         RevUTREngine, EvaluationEngine, WhisperXEngine, NemoSpeakerEngine,
                         StanzaUtteranceEngine, CorefEngine, Wave2VecFAEngine, SeamlessTranslationModel, TencentEngine,
-                        GoogleTranslateEngine, OAIWhisperEngine)
+                        GoogleTranslateEngine, OAIWhisperEngine, TencentUTREngine)
 
 from batchalign import BatchalignPipeline
 from batchalign.models import resolve
@@ -136,6 +136,8 @@ def dispatch_pipeline(pkg_str, lang, num_speakers=None, **arg_overrides):
             engines.append(SeamlessTranslationModel())
         elif engine == "tencent":
             engines.append(TencentEngine(lang=lang))
+        elif engine == "tencent_utr":
+            engines.append(TencentUTREngine(lang=lang))
         elif engine == "gtrans":
             engines.append(GoogleTranslateEngine())
         elif engine == "whisper_oai":

{batchalignhk-0.7.19.post13 → batchalignhk-0.7.19.post15}/batchalign/pipelines/utr/__init__.py

@@ -1,2 +1,3 @@
 from .whisper_utr import WhisperUTREngine
 from .rev_utr import RevUTREngine
+from .tencent_utr import TencentUTREngine

batchalignhk-0.7.19.post15/batchalign/pipelines/utr/tencent_utr.py

@@ -0,0 +1,230 @@
+"""
+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.pipelines.utr.utils import bulletize_doc
+
+from batchalign.errors import *
+import warnings 
+
+import time
+import pathlib
+import pycountry
+
+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
+
+
+import logging
+L = logging.getLogger("batchalign")
+
+class TencentUTREngine(BatchalignEngine):
+    tasks = [ Task.UTTERANCE_TIMING_RECOVERY ]
+
+    def __init__(self, key:str=None, lang="eng"):
+
+        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
+
+        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")
+
+
+    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)
+
+    @staticmethod
+    def is_roman(x):
+        """check if x contains only roman characters"""
+        return all(c.isalpha() and ord(c) < 128 for c in x if not c.isspace())
+    
+
+
+    def process(self, doc, **kwargs):
+        # bring language code into the stack to access
+        lang = self.__lang
+        try:
+            lang = pycountry.languages.get(alpha_3=doc.langs[0]).alpha_2
+        except:
+            # some languages don't have alpha 2
+            pass
+
+        if lang == "zh":
+            lang = "cmn"
+
+        client = self.__client
+
+
+        # check and if there are existing utterance timings, warn
+        if any([i.alignment for i in doc.content if isinstance(i, Utterance)]):
+            warnings.warn(f"We found existing utterance timings in the document with {doc.media.url}! Skipping rough utterance alignment.")
+            return doc
+
+        f = kwargs.get("extra_info", {}).get("extra_input")
+
+        if not f:
+            assert doc.media != None and doc.media.url != None, f"We cannot add utterance timings to something that doesn't have a media path! Provided media tier='{doc.media}'"
+
+        f = f if f else doc.media.url
+
+        lang = self.__lang
+        client = self.__client
+
+
+
+        # processed_path = self.__preprocess_audio(f)
+        # audio = AudioSegment.from_file(processed_path)
+        
+        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()
+        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
+        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 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
+            roman_cache = ""
+            roman_cache_start = i.StartMs
+            roman_cache_end = i.StartMs
+            for j in i.Words:
+                word = j.Word
+                if self.__lang == "yue":
+                    word = cc.convert(word)
+
+                    word = self.replace_cantonese_words(word)
+
+                if self.is_roman(word):
+                    if roman_cache == "":
+                        roman_cache_start = (j.OffsetStartMs + start)
+                    roman_cache = roman_cache + word
+                    roman_cache_end = (j.OffsetEndMs + start)
+                else:
+                    if roman_cache != "":
+                        turn.append({
+                            "type": "text",
+                            "ts": roman_cache_start / 1000,
+                            "end_ts": roman_cache_end / 1000,
+                            "value": roman_cache
+                        })
+                    roman_cache = ""
+                    turn.append({
+                        "type": "text",
+                        "ts": (j.OffsetStartMs + start) / 1000,
+                        "end_ts": (j.OffsetEndMs + start) / 1000,
+                        "value": word
+                    })
+
+            if roman_cache != "":
+                turn.append({
+                    "type": "text",
+                    "ts": roman_cache_start / 1000,
+                    "end_ts": roman_cache_end / 1000,
+                    "value": roman_cache
+                })
+
+            turns.append({
+                "elements": turn,
+                "speaker": i.SpeakerId
+            })
+        L.debug(f"Tencent done.")
+
+        res =  bulletize_doc({"monologues": turns}, doc)
+        L.debug(f"done...")
+
+        return res
+

batchalignhk-0.7.19.post15/batchalign/version

@@ -0,0 +1,3 @@
+0.7.19-post.15
+June 11th, 2025
+wip support for tencent in FA

batchalignhk-0.7.19.post13/batchalign/version

@@ -1,3 +0,0 @@
-0.7.19-post.13
-May 27th, 2025
---data flag