batchalign 0.7.13__tar.gz → 0.7.14__tar.gz

{batchalign-0.7.13/batchalign.egg-info → batchalign-0.7.14}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.13
+Version: 0.7.14
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu
@@ -33,6 +33,9 @@ Requires-Dist: soundfile~=0.12.0
 Requires-Dist: rich-click>=1.7.0
 Requires-Dist: typing-extensions
 Requires-Dist: num2words
+Requires-Dist: tiktoken
+Requires-Dist: blobfile
+Requires-Dist: sentencepiece
 Provides-Extra: dev
 Requires-Dist: pytest; extra == "dev"
 Provides-Extra: train

{batchalign-0.7.13 → batchalign-0.7.14}/batchalign/cli/cli.py

@@ -196,6 +196,28 @@ def transcribe(ctx, in_dir, out_dir, lang, num_speakers, **kwargs):
                   loader, writer, C,
                   asr=asr, **kwargs)
 
+#################### TRANSLATE ################################
+
+@batchalign.command()
+@common_options
+@click.pass_context
+def translate(ctx, in_dir, out_dir, **kwargs):
+    """Translate the transcript to English."""
+
+    def loader(file):
+        cf = CHATFile(path=os.path.abspath(file), special_mor_=True)
+        doc = cf.doc
+        # if str(cf).count("%mor") > 0:
+        #     doc.ba_special_["special_mor_notation"] = True
+        return doc
+
+    def writer(doc, output):
+        CHATFile(doc=doc).write(output)
+
+    _dispatch("translate", "eng", 1, ["cha"], ctx,
+              in_dir, out_dir,
+              loader, writer, C)
+
 #################### MORPHOTAG ################################
 
 @batchalign.command()

{batchalign-0.7.13 → batchalign-0.7.14}/batchalign/cli/dispatch.py

@@ -48,6 +48,7 @@ Cmd2Task = {
     "benchmark": "asr,eval",
     "utseg": "utterance",
     "coref": "coref",
+    "translate": "translate",
 }
 
 # this is the main runner used by all functions

{batchalign-0.7.13 → batchalign-0.7.14}/batchalign/document.py

@@ -31,6 +31,7 @@ class Task(IntEnum):
     MORPHOSYNTAX = 11
     COREF = 12
     WER = 13
+    TRANSLATE = 14
 
 
     DEBUG__G = 0
@@ -54,6 +55,7 @@ TypeMap = {
     Task.DISFLUENCY_ANALYSIS: TaskType.PROCESSING,
     Task.COREF: TaskType.PROCESSING,
     Task.WER: TaskType.ANALYSIS,
+    Task.TRANSLATE: TaskType.PROCESSING,
 
     Task.DEBUG__G: TaskType.GENERATION,
     Task.DEBUG__P: TaskType.PROCESSING,
@@ -73,6 +75,7 @@ TaskFriendlyName = {
     Task.DISFLUENCY_ANALYSIS:  "Disfluncy Analysis",
     Task.COREF:  "Coreference Resolution",
     Task.WER:  "Word Error Rate",
+    Task.TRANSLATE:  "Translation",
     Task.DEBUG__G:  "TEST_GENERATION",
     Task.DEBUG__P:  "TEST_PROCESSING",
     Task.DEBUG__A:   "TEST_ANALYSIS",
@@ -150,6 +153,7 @@ class Utterance(BaseModel):
     tier: Tier = Field(default=Tier())
     content: Sentence
     text: Optional[str] = Field(default=None)
+    translation: Optional[str] = Field(default=None)
     time: Optional[Tuple[int,int]] = Field(default=None)
     custom_dependencies: List[CustomLine]  = Field(default=[])
 

{batchalign-0.7.13 → batchalign-0.7.14}/batchalign/formats/chat/generator.py

@@ -95,7 +95,8 @@ def generate_chat_utterance(utterance: Utterance, special_mor=False, write_wor=T
         result.append("%wor:\t"+" ".join(wor_elems))
     if has_coref:
         result.append("%coref:\t"+" ".join(coref_elems))
-
+    if utterance.translation != None:
+        result.append("%xtra:\t"+utterance.translation)
 
 
     #### EXTRA LINE GENERATION ####

{batchalign-0.7.13 → batchalign-0.7.14}/batchalign/formats/chat/parser.py

@@ -280,6 +280,7 @@ def chat_parse_doc(lines, special_mor=False):
                 mor = None
                 gra = None
                 wor = None
+                translation = None
                 additional = []
 
                 while raw[0][0] == "%":
@@ -291,6 +292,8 @@ def chat_parse_doc(lines, special_mor=False):
                         gra = line
                     elif beg.strip() == "wor" or beg.strip() == "xwor":
                         wor = line
+                    elif beg.strip() == "xtra":
+                        translation = line
                     else:
                         additional.append(CustomLine(id=beg.strip(),
                                                         type=CustomLineType.DEPENDENT,
@@ -309,7 +312,8 @@ def chat_parse_doc(lines, special_mor=False):
                     "content": parsed,
                     "text": text,
                     "delim": delim,
-                    "custom_dependencies": additional
+                    "custom_dependencies": additional,
+                    "translation": translation
                 })
 
                 timing = re.findall(rf"\x15(\d+)_(\d+)\x15", text)

{batchalign-0.7.13 → batchalign-0.7.14}/batchalign/pipelines/__init__.py

@@ -12,3 +12,4 @@ from .utr import WhisperUTREngine, RevUTREngine
 from .analysis import EvaluationEngine
 from .utterance import StanzaUtteranceEngine
 
+from .translate import SeamlessTranslationModel

{batchalign-0.7.13 → batchalign-0.7.14}/batchalign/pipelines/dispatch.py

@@ -6,7 +6,7 @@ Tabulate default packages and options.
 from batchalign import (WhisperEngine, WhisperFAEngine, StanzaEngine, RevEngine,
                         NgramRetraceEngine, DisfluencyReplacementEngine, WhisperUTREngine,
                         RevUTREngine, EvaluationEngine, WhisperXEngine, NemoSpeakerEngine,
-                        StanzaUtteranceEngine, CorefEngine, Wave2VecFAEngine)
+                        StanzaUtteranceEngine, CorefEngine, Wave2VecFAEngine, SeamlessTranslationModel)
 from batchalign import BatchalignPipeline
 from batchalign.models import resolve
 
@@ -28,6 +28,7 @@ DEFAULT_PACKAGES = {
     "eval": "evaluation",
     "utterance": "stanza_utt",
     "coref": "stanza_coref",
+    "translate": "seamless_translate",
 }
 
 LANGUAGE_OVERRIDE_PACKAGES = {
@@ -129,6 +130,8 @@ def dispatch_pipeline(pkg_str, lang, num_speakers=None, **arg_overrides):
             engines.append(CorefEngine())
         elif engine == "wav2vec_fa":
             engines.append(Wave2VecFAEngine())
+        elif engine == "seamless_translate":
+            engines.append(SeamlessTranslationModel())
 
     L.debug(f"Done initalizing packages.")
     return BatchalignPipeline(*engines)

{batchalign-0.7.13 → batchalign-0.7.14}/batchalign/pipelines/morphosyntax/ud.py

@@ -826,6 +826,7 @@ def morphoanalyze(doc: Document, retokenize:bool, status_hook:callable = None, *
         line_cut = line_cut.replace("+//", "")
         line_cut = line_cut.replace("+...", "")
         line_cut = line_cut.replace("_", "")
+        line_cut = line_cut.replace("#", "")
 
         # xbxxx is a sepecial xxx-class token to mark
         # special form markers, used for processing later

batchalign-0.7.14/batchalign/pipelines/translate/__init__.py

@@ -0,0 +1 @@
+from .seamless import SeamlessTranslationModel

batchalign-0.7.14/batchalign/pipelines/translate/seamless.py

@@ -0,0 +1,53 @@
+from batchalign.models import WhisperFAModel
+from batchalign.document import *
+from batchalign.pipelines.base import *
+from batchalign.utils import *
+from batchalign.utils.dp import *
+from batchalign.constants import *
+
+from transformers import AutoProcessor, SeamlessM4TModel
+
+import logging
+L = logging.getLogger("batchalign")
+
+import re
+
+# !uv pip install sentencepiece
+
+import pycountry
+import warnings
+
+class SeamlessTranslationModel(BatchalignEngine):
+    tasks = [ Task.TRANSLATE ]
+
+    def _hook_status(self, status_hook):
+        self.status_hook = status_hook
+
+    def __init__(self):
+        self.status_hook = None
+        self.processor = AutoProcessor.from_pretrained("facebook/hf-seamless-m4t-medium")
+        self.model = SeamlessM4TModel.from_pretrained("facebook/hf-seamless-m4t-medium")
+
+    def process(self, doc:Document, **kwargs):
+
+        for indx, i in enumerate(doc.content):
+            if not isinstance(i, Utterance):
+                continue
+            if i.translation:
+                continue
+            
+            text = i.strip(join_with_spaces=False, include_retrace=True, include_fp=True)
+            text_inputs = self.processor(text=text, src_lang=doc.langs[0] if doc.langs[0] != "zho" else "cmn", return_tensors="pt")
+            output_tokens = self.model.generate(**text_inputs, tgt_lang="eng", generate_speech=False)
+            translated_text_from_text = self.processor.decode(output_tokens[0].tolist()[0], skip_special_tokens=True)
+
+            i.translation = translated_text_from_text
+            for j in MOR_PUNCT + ENDING_PUNCT:
+                i.translation = i.translation.replace(j, " "+j)
+
+            if self.status_hook != None:
+                self.status_hook(indx+1, len(doc.content))
+
+        return doc
+
+

batchalign-0.7.14/batchalign/version

@@ -0,0 +1,3 @@
+0.7.14
+Feburary 19nd, 2025
+machine translation!

{batchalign-0.7.13 → batchalign-0.7.14/batchalign.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.13
+Version: 0.7.14
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu
@@ -33,6 +33,9 @@ Requires-Dist: soundfile~=0.12.0
 Requires-Dist: rich-click>=1.7.0
 Requires-Dist: typing-extensions
 Requires-Dist: num2words
+Requires-Dist: tiktoken
+Requires-Dist: blobfile
+Requires-Dist: sentencepiece
 Provides-Extra: dev
 Requires-Dist: pytest; extra == "dev"
 Provides-Extra: train

{batchalign-0.7.13 → batchalign-0.7.14}/batchalign.egg-info/SOURCES.txt

@@ -83,6 +83,8 @@ batchalign/pipelines/morphosyntax/fr/case.py
 batchalign/pipelines/morphosyntax/ja/verbforms.py
 batchalign/pipelines/speaker/__init__.py
 batchalign/pipelines/speaker/nemo_speaker.py
+batchalign/pipelines/translate/__init__.py
+batchalign/pipelines/translate/seamless.py
 batchalign/pipelines/utr/__init__.py
 batchalign/pipelines/utr/rev_utr.py
 batchalign/pipelines/utr/utils.py

{batchalign-0.7.13 → batchalign-0.7.14}/batchalign.egg-info/requires.txt

@@ -23,6 +23,9 @@ soundfile~=0.12.0
 rich-click>=1.7.0
 typing-extensions
 num2words
+tiktoken
+blobfile
+sentencepiece
 
 [dev]
 pytest

{batchalign-0.7.13 → batchalign-0.7.14}/setup.py

@@ -52,6 +52,9 @@ setup(
         "rich-click>=1.7.0",
         "typing-extensions",
         "num2words",
+        "tiktoken",
+        "blobfile",
+        "sentencepiece"
     ],
     extras_require={
         'dev': [

batchalign-0.7.13/batchalign/version

@@ -1,3 +0,0 @@
-0.7.13
-Feburary 6nd, 2025
-numbers support