BatchalignHK 0.7.23.post1__tar.gz → 0.8.0.post1__tar.gz

{batchalignhk-0.7.23.post1 → batchalignhk-0.8.0.post1}/BatchalignHK.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: BatchalignHK
-Version: 0.7.23.post1
+Version: 0.8.0.post1
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

{batchalignhk-0.7.23.post1 → batchalignhk-0.8.0.post1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: BatchalignHK
-Version: 0.7.23.post1
+Version: 0.8.0.post1
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

{batchalignhk-0.7.23.post1 → batchalignhk-0.8.0.post1}/batchalign/cli/cli.py

@@ -3,37 +3,23 @@ cli.py
 The Batchalign command-line interface
 """
 
-import multiprocessing
 import rich_click as click
 import functools
 
 import os
-from glob import glob
 
-from multiprocessing import Process, freeze_support
-
-from batchalign.pipelines import BatchalignPipeline
+from multiprocessing import freeze_support
 
+from pathlib import Path
 from rich.traceback import install
 from rich.console import Console
-from rich.panel import Panel
-from pathlib import Path
-from batchalign.document import *
-from batchalign.formats.chat import CHATFile
-from batchalign.utils import config
 from rich.logging import RichHandler
 
 from batchalign.cli.dispatch import _dispatch
 from batchalign.models.training.run import cli as train
 
-from enum import Enum
-
-import traceback
-
 import pyfiglet
-from rich import pretty
-import logging as L 
-baL = L.getLogger('batchalign')
+import logging as L
 
 C = Console()
 
@@ -62,7 +48,7 @@ def handle_verbosity(verbosity):
     L.getLogger('stanza').handlers.clear()
     L.getLogger('transformers').handlers.clear()
     L.getLogger('nemo_logger').handlers.clear()
-    L.getLogger("stanza").setLevel(L.INFO)
+    L.getLogger("stanza").setLevel(L.WARN)
     L.getLogger('nemo_logger').setLevel(L.CRITICAL)
     L.getLogger('batchalign').setLevel(L.WARN)
     L.getLogger('lightning.pytorch.utilities.migration.utils').setLevel(L.ERROR)
@@ -73,6 +59,7 @@ def handle_verbosity(verbosity):
         L.getLogger('batchalign').setLevel(L.INFO)
     if verbosity >= 3:
         L.getLogger('batchalign').setLevel(L.DEBUG)
+        L.getLogger("stanza").setLevel(L.INFO)
     if verbosity >= 4:
         L.getLogger('batchalign').setLevel(L.DEBUG)
         L.getLogger('transformers').setLevel(L.INFO)
@@ -81,7 +68,8 @@ def handle_verbosity(verbosity):
 @click.pass_context
 @click.version_option(VERSION_NUMBER)
 @click.option("-v", "--verbose", type=int, count=True, default=0, help="How loquacious Batchalign should be.")
-def batchalign(ctx, verbose):
+@click.option("--workers", type=int, default=os.cpu_count(), help="Number of worker processes to use.")
+def batchalign(ctx, verbose, workers):
     """process .cha and/or audio files in IN_DIR and dumps them to OUT_DIR using recipe COMMAND"""
 
     ## setup commands ##
@@ -93,7 +81,9 @@ def batchalign(ctx, verbose):
     handle_verbosity(verbose)
     # add to arguments
     ctx.obj["verbose"] = verbose
+    ctx.obj["workers"] = workers
     # setup config
+    from batchalign.utils import config
     ctx.obj["config"] = config.config_read(True)
     # make everything look better
     # pretty.install()
@@ -122,6 +112,7 @@ batchalign.add_command(train, "models")
 @click.pass_context
 def align(ctx, in_dir, out_dir, whisper, wav2vec, iic, wav2vec_yue, tencent, funaudio, **kwargs):
     """Align transcripts against corresponding media files."""
+    from batchalign.formats.chat import CHATFile
     def loader(file):
         return (
             CHATFile(path=os.path.abspath(file)).doc,
@@ -180,6 +171,8 @@ def align(ctx, in_dir, out_dir, whisper, wav2vec, iic, wav2vec_yue, tencent, fun
 @click.pass_context
 def transcribe(ctx, in_dir, out_dir, lang, num_speakers, **kwargs):
     """Create a transcript from audio files."""
+    from batchalign.document import CustomLine, CustomLineType
+    from batchalign.formats.chat import CHATFile
     def loader(file):
         return file
 
@@ -229,6 +222,7 @@ def transcribe(ctx, in_dir, out_dir, lang, num_speakers, **kwargs):
 @click.pass_context
 def translate(ctx, in_dir, out_dir, **kwargs):
     """Translate the transcript to English."""
+    from batchalign.formats.chat import CHATFile
 
     def loader(file):
         cf = CHATFile(path=os.path.abspath(file), special_mor_=True)
@@ -259,6 +253,7 @@ def translate(ctx, in_dir, out_dir, **kwargs):
 @click.pass_context
 def morphotag(ctx, in_dir, out_dir, **kwargs):
     """Perform morphosyntactic analysis on transcripts."""
+    from batchalign.formats.chat import CHATFile
 
     def loader(file):
         mwt = {}
@@ -285,7 +280,7 @@ def morphotag(ctx, in_dir, out_dir, **kwargs):
 
     _dispatch("morphotag", "eng", 1, ["cha"], ctx,
               in_dir, out_dir,
-              loader, writer, C)
+              loader, writer, C, **kwargs)
 
 
 #################### MORPHOTAG ################################
@@ -295,6 +290,7 @@ def morphotag(ctx, in_dir, out_dir, **kwargs):
 @click.pass_context
 def coref(ctx, in_dir, out_dir, **kwargs):
     """Perform coreference analysis on transcripts."""
+    from batchalign.formats.chat import CHATFile
 
     def loader(file):
         cf = CHATFile(path=os.path.abspath(file))
@@ -322,6 +318,7 @@ def coref(ctx, in_dir, out_dir, **kwargs):
 @click.pass_context
 def utseg(ctx, in_dir, out_dir, lang, num_speakers, **kwargs):
     """Perform morphosyntactic analysis on transcripts."""
+    from batchalign.formats.chat import CHATFile
 
     def loader(file):
         return CHATFile(path=os.path.abspath(file)).doc
@@ -354,6 +351,7 @@ def utseg(ctx, in_dir, out_dir, lang, num_speakers, **kwargs):
 @click.pass_context
 def benchmark(ctx, in_dir, out_dir, lang, num_speakers, whisper, tencent, funaudio, whisper_oai, **kwargs):
     """Benchmark ASR utilities for their word accuracy"""
+    from batchalign.formats.chat import CHATFile
     def loader(file):
         # try to find a .cha in the same directory
         p = Path(file)
@@ -397,6 +395,7 @@ def avqi(ctx, input_dir, output_dir, lang, **kwargs):
     """Calculate AVQI from paired .cs and .sv audio files in input directory."""
 
     from batchalign.pipelines.avqi import AVQIEngine
+    from batchalign.document import Document
     from pathlib import Path
     import os
     
@@ -464,6 +463,7 @@ def avqi(ctx, input_dir, output_dir, lang, **kwargs):
 @click.pass_context
 def opensmile(ctx, input_dir, output_dir, feature_set, lang, **kwargs):
     """Extract openSMILE audio features from speech samples."""
+    from batchalign.document import Document
 
     def loader(file):
         doc = Document.new(media_path=file, lang=lang)
@@ -491,6 +491,7 @@ def opensmile(ctx, input_dir, output_dir, feature_set, lang, **kwargs):
 def setup(ctx):
     """Reconfigure Batchalign settings, such as Rev.AI key."""
 
+    from batchalign.utils import config
     config.interactive_setup()
 
 #################### VERSION ################################
@@ -503,5 +504,5 @@ def version(ctx, **kwargs):
     ptr = (pyfiglet.figlet_format("Batchalign2")+"\n" +
            f"Version: [bold]{VERSION_NUMBER.strip()}[/bold], released {RELEASE_DATE.strip()}\n" +
            f"[italic]{RELEASE_NOTES.strip()}[/italic]"+"\n" +
-           "\nDeveloped by Brian MacWhinney and Houjun Liu")
+           "\nDeveloped by Brian MacWhinney and Houjun Liu\ncontributions from Sebastian Song and Franklin Chen")
     C.print("\n\n"+ptr+"\n\n")

batchalignhk-0.8.0.post1/batchalign/cli/dispatch.py

@@ -0,0 +1,419 @@
+"""
+dispatch.py
+CLI runner dispatch. Essentially the translation layer between `command` in CLI
+and actual BatchalignPipeline.
+"""
+
+from rich.progress import Progress, SpinnerColumn, TextColumn, TimeElapsedColumn, BarColumn
+from urllib.parse import urlparse
+
+import warnings
+
+import shutil
+import os
+import glob
+import queue
+
+from rich.console import Console
+from rich.markup import escape
+
+from pathlib import Path
+
+import concurrent.futures
+import multiprocessing
+from functools import partial
+
+# Oneliner of directory-based glob and replace
+globase = lambda path, statement: glob(os.path.join(path, statement))
+repath_file = lambda file_path, new_dir: os.path.join(new_dir, Path(file_path).name)
+
+import tempfile
+import time
+
+import traceback
+import logging as L
+baL = L.getLogger('batchalign')
+
+warnings.filterwarnings('ignore', category=UserWarning, message='TypedStorage is deprecated')
+
+# Global cache for the pipeline in worker processes
+_worker_pipeline = None
+
+def _get_worker_pipeline(command, lang, num_speakers, **kwargs):
+    global _worker_pipeline
+    if _worker_pipeline is None:
+        from batchalign.pipelines import BatchalignPipeline
+        _worker_pipeline = BatchalignPipeline.new(Cmd2Task[command],
+                                                lang=lang, num_speakers=num_speakers, **kwargs)
+    return _worker_pipeline
+
+def _worker_task(file_info, command, lang, num_speakers, loader_info, writer_info, progress_queue=None, verbose=0, **kwargs):
+    """The task executed in each worker process."""
+    import sys
+    import os
+    import tempfile
+    import logging
+
+    file, output = file_info
+    pid = os.getpid()
+
+    # Configure logging in this worker process
+    if verbose >= 1:
+        # Ensure basicConfig is called so logging works
+        logging.basicConfig(format="%(levelname)s - %(name)s - %(message)s", level=logging.ERROR)
+
+    # Configure batchalign logger level in this worker process
+    baL = logging.getLogger('batchalign')
+    if verbose == 0:
+        baL.setLevel(logging.WARN)
+    elif verbose == 1:
+        baL.setLevel(logging.INFO)
+    else:
+        baL.setLevel(logging.DEBUG)
+
+    # Only capture output if not in verbose mode
+    # In verbose mode, let logs stream naturally to the console
+    should_capture = verbose == 0
+
+    if should_capture:
+        # Use a temporary file to capture ALL output at the FD level
+        # This is the most robust way to prevent interleaved output
+        log_file = tempfile.TemporaryFile(mode='w+')
+        old_stdout_fd = os.dup(sys.stdout.fileno())
+        old_stderr_fd = os.dup(sys.stderr.fileno())
+
+        # Redirect FD 1 and 2 to our temp file
+        os.dup2(log_file.fileno(), sys.stdout.fileno())
+        os.dup2(log_file.fileno(), sys.stderr.fileno())
+
+    try:
+        pipeline = _get_worker_pipeline(command, lang, num_speakers, **kwargs)
+
+        def progress_callback(completed, total, tasks):
+            if not progress_queue:
+                return
+            try:
+                progress_queue.put((file, completed, total, tasks))
+            except Exception:
+                pass
+
+        # For now, we'll re-import what we need
+        from batchalign.formats.chat import CHATFile
+
+        # Morphosyntax specific loader/writer logic moved here for picklability
+        if command == "morphotag":
+            # Extract morphotag-specific arguments from kwargs
+            mwt = kwargs.pop("mwt", {})
+            retokenize = kwargs.pop("retokenize", False)
+            skipmultilang = kwargs.pop("skipmultilang", False)
+
+            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
+
+            # Prepare arguments for the pipeline
+            pipeline_kwargs = {
+                "retokenize": retokenize,
+                "skipmultilang": skipmultilang,
+                "mwt": mwt
+            }
+            # Add any remaining kwargs
+            pipeline_kwargs.update(kwargs)
+
+            # Process
+            doc = pipeline(doc, callback=progress_callback, **pipeline_kwargs)
+
+            # Write
+            CHATFile(doc=doc, special_mor_=doc.ba_special_.get("special_mor_notation", False)).write(output)
+
+        # Add other commands as needed, or use a more generic registry
+        elif command == "align":
+            cf = CHATFile(path=os.path.abspath(file))
+            doc = cf.doc
+            kw = {"pauses": kwargs.get("pauses", False)}
+            doc = pipeline(doc, callback=progress_callback, **kw)
+            CHATFile(doc=doc).write(output, write_wor=kwargs.get("wor", True))
+
+        else:
+            loader, writer = loader_info, writer_info
+            doc = loader(os.path.abspath(file))
+            kw = {}
+            if isinstance(doc, tuple) and len(doc) > 1:
+                doc, kw = doc
+            doc = pipeline(doc, callback=progress_callback, **kw)
+            writer(doc, output)
+
+        # Flush and read captured output if we were capturing
+        if should_capture:
+            sys.stdout.flush()
+            sys.stderr.flush()
+            log_file.seek(0)
+            captured = log_file.read()
+        else:
+            captured = ""
+
+        return file, None, None, captured
+    except Exception as e:
+        # Flush and read captured output if we were capturing
+        if should_capture:
+            sys.stdout.flush()
+            sys.stderr.flush()
+            log_file.seek(0)
+            captured = log_file.read()
+        else:
+            captured = ""
+        return file, traceback.format_exc(), e, captured
+    finally:
+        # Restore original FDs only if we redirected them
+        if should_capture:
+            os.dup2(old_stdout_fd, sys.stdout.fileno())
+            os.dup2(old_stderr_fd, sys.stderr.fileno())
+            os.close(old_stdout_fd)
+            os.close(old_stderr_fd)
+            log_file.close()
+
+# this dictionary maps what commands are executed
+# against what BatchalignPipeline tasks are actually ran 
+Cmd2Task = {
+    "align": "fa",
+    "transcribe": "asr",
+    "transcribe_s": "asr,speaker",
+    "morphotag": "morphosyntax",
+    "benchmark": "asr,eval",
+    "utseg": "utterance",
+    "coref": "coref",
+    "translate": "translate",
+    "opensmile": "opensmile",
+}
+
+# this is the main runner used by all functions
+def _dispatch(command, lang, num_speakers,
+              extensions, ctx, in_dir, out_dir,
+              loader:callable, writer:callable, console,
+              **kwargs):
+
+    C = console
+    from batchalign.constants import FORCED_CONVERSION
+    from batchalign.document import TaskFriendlyName
+
+    # get files by walking the directory
+    files = []
+    outputs = []
+
+    if kwargs.get("data"):
+        url = kwargs.get("data")
+        with open(url.strip()) as data:
+            data = data.readlines()
+        data = [i.strip() for i in data if i.strip() != ""]
+        for url in data:
+            url = urlparse(url)
+            if url.scheme == "":
+                url = url._replace(scheme="http")
+            base = os.path.basename(url.path)
+            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))
+            ext = path.suffix.strip(".").strip().lower()
+
+            # calculate input path, convert if needed
+            inp_path = str(path)
+            if ext in FORCED_CONVERSION:
+                # check for ffmpeg
+                if not shutil.which("ffmpeg"):
+                    raise ValueError(f"ffmpeg not found in Path! Cannot load input media at {inp_path}.\nHint: Please convert your input audio sample to .wav before proceeding witch Batchalign, or install ffmpeg (https://ffmpeg.org/download.html)")
+                # convert
+                from pydub import AudioSegment
+                seg = AudioSegment.from_file(inp_path, ext)
+                seg.export(inp_path.replace(f".{ext}", ".wav"), format="wav")
+                inp_path = inp_path.replace(f".{ext}", ".wav")
+
+            # repath the file to the output
+            rel = os.path.relpath(inp_path, in_dir)
+            repathed = Path(os.path.join(out_dir, rel))
+            # make the repathed dir, if it doesn't exist
+            parent = repathed.parent.absolute()
+            os.makedirs(parent, exist_ok=True)
+
+            # HACK check for @Options:\tdummy in the file
+            # and simply copy it
+            if ext == "cha":
+                with open(inp_path, 'r', encoding="utf-8") as df:
+                    data = df.read()
+                if "@Options:\tdummy" in data:
+                    shutil.copy2(inp_path, str(repathed))
+                    continue
+                elif "This is a dummy file to permit playback from the TalkBank browser" in data:
+                    shutil.copy2(inp_path, str(repathed))
+                    continue
+                
+            # 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
+            else:
+                shutil.copy2(inp_path, str(repathed))
+
+    __tf = None
+    # output file
+    if ctx.obj["verbose"] > 1:
+        __tf = tempfile.NamedTemporaryFile(delete=True, mode='w')
+        C = Console(file=__tf)
+
+    # process largest inputs first to avoid late stragglers
+    file_pairs = list(zip(files, outputs))
+    file_pairs.sort(key=lambda fo: os.path.getsize(fo[0]) if os.path.exists(fo[0]) else 0, reverse=True)
+    files, outputs = zip(*file_pairs) if file_pairs else ([], [])
+
+    C.print(f"\nMode: [blue]{command}[/blue]; got [bold cyan]{len(files)}[/bold cyan] transcript{'s' if len(files) > 1 else ''} to process from {in_dir}:\n")
+
+    # Determine number of workers
+    num_workers = kwargs.get("num_workers", ctx.obj.get("workers", os.cpu_count()))
+
+    # Pre-download stanza resources if needed to avoid interleaved downloads in workers
+    if command in ["morphotag", "utseg", "coref"]:
+        try:
+            import stanza
+            stanza.download_resources_json()
+        except Exception:
+            pass
+
+    # For some commands or environments, we might want to limit this
+    if command in ["transcribe", "transcribe_s"]:
+        num_workers = min(num_workers, 2) # GPU memory limits
+
+    C.print(f"Using [bold]{num_workers}[/bold] worker processes.\n")
+
+    manager = multiprocessing.Manager() if files else None
+    progress_queue = manager.Queue() if manager else None
+
+    def render_stage(stage_tasks):
+        if not stage_tasks:
+            return "Processing..."
+        if not isinstance(stage_tasks, (list, tuple)):
+            stage_tasks = [stage_tasks]
+        names = [TaskFriendlyName.get(task, str(task)) for task in stage_tasks]
+        return ", ".join(names)
+
+    # create the spinner
+    prog = Progress(SpinnerColumn(), *Progress.get_default_columns()[:-1],
+                    TimeElapsedColumn(),
+                    TextColumn("[cyan]{task.fields[processor]}[/cyan]"), console=C)
+    errors = []
+
+    try:
+        with prog as prog:
+            tasks = {}
+            task_totals = {}
+
+            for f in files:
+                tasks[f] = prog.add_task(Path(f).name, start=False, total=1, processor="Waiting...")
+                task_totals[f] = 1
+
+            def drain_progress_queue():
+                if not progress_queue:
+                    return
+                while True:
+                    try:
+                        file, completed, total, stage_tasks = progress_queue.get_nowait()
+                    except queue.Empty:
+                        break
+                    except Exception:
+                        break
+                    if file not in tasks:
+                        continue
+                    task_total = max(int(total) if total else task_totals.get(file, 1), 1)
+                    task_totals[file] = task_total
+                    prog.update(tasks[file],
+                                total=task_total,
+                                completed=min(int(completed), task_total),
+                                processor=render_stage(stage_tasks))
+
+            with concurrent.futures.ProcessPoolExecutor(max_workers=num_workers) as executor:
+                worker_func = partial(_worker_task,
+                                      command=command,
+                                      lang=lang,
+                                      num_speakers=num_speakers,
+                                      loader_info=None,
+                                      writer_info=None,
+                                      progress_queue=progress_queue,
+                                      verbose=ctx.obj["verbose"],
+                                      **kwargs)
+
+                future_to_file = {executor.submit(worker_func, (f, o)): f for f, o in zip(files, outputs)}
+
+                for f in files:
+                    prog.start_task(tasks[f])
+                    prog.update(tasks[f], processor="Processing...")
+
+                pending = set(future_to_file.keys())
+                while pending:
+                    done, pending = concurrent.futures.wait(
+                        pending,
+                        timeout=0.1,
+                        return_when=concurrent.futures.FIRST_COMPLETED,
+                    )
+                    drain_progress_queue()
+
+                    for future in done:
+                        file = future_to_file[future]
+                        try:
+                            res_file, trcbk, e, captured = future.result()
+                            final_total = max(task_totals.get(file, 1), 1)
+                            if e:
+                                prog.update(tasks[file], total=final_total, completed=final_total, processor="[bold red]FAIL[/bold red]")
+                                errors.append((res_file, trcbk, e, captured))
+                            else:
+                                prog.update(tasks[file], total=final_total, completed=final_total, processor="[bold green]DONE[/bold green]")
+                                if ctx.obj["verbose"] >= 1 and captured.strip():
+                                    errors.append((res_file, "Logs only (Success)", None, captured))
+                        except Exception as e:
+                            final_total = max(task_totals.get(file, 1), 1)
+                            prog.update(tasks[file], total=final_total, completed=final_total, processor="[bold red]FAIL[/bold red]")
+                            errors.append((file, traceback.format_exc(), e, ""))
+
+                drain_progress_queue()
+    finally:
+        if manager:
+            manager.shutdown()
+
+    if len(errors) > 0:
+        C.print()
+        for file, trcbk, e, captured in errors:
+            rel_path = os.path.relpath(str(Path(file).absolute()), in_dir)
+            if e:
+                C.print(f"[bold red]ERROR[/bold red] on file [italic]{rel_path}[/italic]: {escape(str(e))}\n")
+                if captured.strip():
+                    C.print(f"[dim]Captured Worker Output:[/dim]\n{escape(captured.strip())}\n")
+                if ctx.obj["verbose"] == 1:
+                    C.print(escape(str(trcbk)))
+                elif ctx.obj["verbose"] > 1:
+                    Console().print(escape(str(trcbk)))
+            elif captured.strip():
+                C.print(f"[bold blue]INFO[/bold blue] on file [italic]{rel_path}[/italic]:\n")
+                C.print(f"{escape(captured.strip())}\n")
+    else:
+        C.print(f"\nAll done. Results saved to {out_dir}!\n")
+
+    if ctx.obj["verbose"] > 1:
+        C.end_capture()
+
+    if __tf:
+        __tf.close()

{batchalignhk-0.7.23.post1 → batchalignhk-0.8.0.post1}/batchalign/formats/chat/generator.py

@@ -42,10 +42,11 @@ def generate_chat_utterance(utterance: Utterance, special_mor=False, write_wor=T
         main_line = re.sub(r"(?:[a-z]) ?\(([a-z]+) ?\)", r"(\1)", main_line)
         main_line = re.sub(r"([a-z]) _", r"\1_", main_line)
         main_line = re.sub(r"  ", r" ", main_line)
+    main_line = re.sub(r"^,", "", main_line.strip()) # remove initial commas
     main_line = re.sub(r"«", "“", main_line)
     main_line = re.sub(r"»", "”", main_line)
     main_line = re.sub(r"—", "-", main_line)
-    main_line = re.sub(r"–", "-", main_line)
+    main_line = re.sub(r"–", "-", main_line).strip()
     tier = utterance.tier
 
     mors = []