lyrics-transcriber 0.17.1__tar.gz → 0.18.0__tar.gz

{lyrics_transcriber-0.17.1 → lyrics_transcriber-0.18.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: lyrics-transcriber
-Version: 0.17.1
+Version: 0.18.0
 Summary: Automatically create synchronised lyrics files in ASS and MidiCo LRC formats with word-level timestamps, using Whisper and lyrics from Genius and Spotify
 Home-page: https://github.com/karaokenerds/python-lyrics-transcriber
 License: MIT
@@ -13,8 +13,8 @@ Classifier: Programming Language :: Python :: 3.9
 Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
 Requires-Dist: Cython (>=0)
-Requires-Dist: auditok (>=0.2)
 Requires-Dist: dtw-python (>=1)
 Requires-Dist: llvmlite (>=0)
 Requires-Dist: lyricsgenius (>=3)

{lyrics_transcriber-0.17.1 → lyrics_transcriber-0.18.0}/lyrics_transcriber/transcriber.py

@@ -308,6 +308,13 @@ class LyricsTranscriber:
                 self.outputs["corrected_lyrics_data_dict"] = corrected_lyrics_data_dict
                 return
 
+        reference_lyrics = self.outputs.get("genius_lyrics_text") or self.outputs.get("spotify_lyrics_text")
+
+        if not reference_lyrics:
+            self.logger.warning("No reference lyrics found from Genius or Spotify. Skipping LLM correction.")
+            self.outputs["corrected_lyrics_data_dict"] = self.outputs["transcription_data_dict"]
+            return
+
         self.logger.debug(
             f"no cached lyrics found at corrected_lyrics_data_json_cache_filepath: {corrected_lyrics_data_json_cache_filepath}, attempting to run correction using LLM"
         )
@@ -317,7 +324,6 @@ class LyricsTranscriber:
         with open(self.llm_prompt_correction, "r") as file:
             system_prompt_template = file.read()
 
-        reference_lyrics = self.outputs["genius_lyrics_text"] or self.outputs["spotify_lyrics_text"]
         system_prompt = system_prompt_template.replace("{{reference_lyrics}}", reference_lyrics)
 
         # TODO: Test if results are cleaner when using the vocal file from a background vocal audio separation model
@@ -639,7 +645,9 @@ class LyricsTranscriber:
                 for i, word in enumerate(segment["words"]):
                     start_time = self.format_time_lrc(word["start"])
                     if i != len(segment["words"]) - 1:
-                        word["text"] += " "
+                        if not word["text"].endswith(" "):
+                            self.logger.debug(f"word '{word['text']}' does not end with a space, adding one")
+                            word["text"] += " "
                     line = "[{}]1:{}{}\n".format(start_time, "/" if i == 0 else "", word["text"])
                     f.write(line)
 
@@ -961,7 +969,10 @@ class LyricsTranscriber:
             self.logger.debug(f"Using Whisper for transcription with model: {self.transcription_model}")
             audio = whisper.load_audio(self.audio_filepath)
             model = whisper.load_model(self.transcription_model, device="cpu")
-            transcription_data = whisper.transcribe(model, audio, language="en", vad="auditok", beam_size=5, temperature=0.2, best_of=5)
+            transcription_data = whisper.transcribe(model, audio, language="en", beam_size=5, temperature=0.2, best_of=5)
+            
+            # auditok is needed for voice activity detection, but it has OS package dependencies that are hard to install on some platforms
+            # transcription_data = whisper.transcribe(model, audio, language="en", vad="auditok", beam_size=5, temperature=0.2, best_of=5)
 
             # Remove segments with no words, only music
             transcription_data["segments"] = [segment for segment in transcription_data["segments"] if segment["text"].strip() != "Music"]

{lyrics_transcriber-0.17.1 → lyrics_transcriber-0.18.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "lyrics-transcriber"
-version = "0.17.1"
+version = "0.18.0"
 description = "Automatically create synchronised lyrics files in ASS and MidiCo LRC formats with word-level timestamps, using Whisper and lyrics from Genius and Spotify"
 authors = ["Andrew Beveridge <andrew@beveridge.uk>"]
 license = "MIT"
@@ -26,13 +26,12 @@ lyricsgenius = ">=3"
 python-slugify = ">=8"
 syrics = ">=0"
 openai = "^1"
-openai-whisper = ">=20231117"
 transformers = ">=4"
-auditok = ">=0.2"
 whisper-timestamped = ">=1"
 tenacity = ">=8"
+openai-whisper = ">=20231117"
 # Note: after adding openai-whisper and whisper-timestamped with poetry lock, I then removed all traces of triton
-# from poetry.lock before running poetry install, as triton doesn't support macOS but isn't actually needed for whisper.
+# from poetry.lock before running poetry install, as triton doesn't support macOS but isn't actually needed for whisper
 # This was the only way I was able to get a working cross-platform build published to PyPI.
 # To update the lockfile and install/upgrade dependencies, modify the dependency list above then run:
 # poetry lock; patch -p0 poetry.lock <.github/removetriton.patch; poetry install