lyrics-transcriber 0.14.0__tar.gz → 0.16.0__tar.gz

{lyrics_transcriber-0.14.0 → lyrics_transcriber-0.16.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: lyrics-transcriber
-Version: 0.14.0
+Version: 0.16.0
 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
@@ -14,6 +14,7 @@ Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Requires-Dist: Cython (>=0)
+Requires-Dist: auditok (>=0.2)
 Requires-Dist: dtw-python (>=1)
 Requires-Dist: llvmlite (>=0)
 Requires-Dist: lyricsgenius (>=3)
@@ -52,7 +53,10 @@ Automatically create synchronised lyrics files in ASS and MidiCo LRC formats wit
 ### Prerequisites
 
 - Python 3.9 or higher
-- [Optional] A Genius API token if you want to fetch lyrics from Genius
+- [Optional] Genius API token if you want to fetch lyrics from Genius
+- [Optional] Spotify cookie value if you want to fetch lyrics from Spotify
+- [Optional] OpenAI API token if you want to use LLM correction of the transcribed lyrics
+- [Optional] AudioShake API token if you want to use a much higher quality (but paid) API for lyrics transcription
 
 ```
 pip install lyrics-transcriber
@@ -61,6 +65,23 @@ pip install lyrics-transcriber
 > **Warning**
 > The package published to PyPI was created by manually editing `poetry.lock` to remove [triton](https://github.com/openai/triton), as it is technically a sub-dependency from openai-whisper but is currently only supported on Linux (whisper still works fine without it, and I want this package to be usable on any platform)
 
+## Docker
+
+You can use the pre-built container image `beveradb/lyrics-transcriber:0.16.0` on Docker hub if you want, here's an example:
+
+```sh
+docker run \
+ -v `pwd`/input:/input \
+ -v `pwd`/output:/output \
+beveradb/lyrics-transcriber:0.16.0 \
+ --log_level debug \
+ --output_dir /output \
+ --render_video \
+ --video_background_image /input/your-background-image.png \
+ --video_resolution 360p \
+ /input/song.flac
+```
+
 ## Usage 🚀
 
 ### As a standalone CLI

{lyrics_transcriber-0.14.0 → lyrics_transcriber-0.16.0}/README.md

@@ -17,7 +17,10 @@ Automatically create synchronised lyrics files in ASS and MidiCo LRC formats wit
 ### Prerequisites
 
 - Python 3.9 or higher
-- [Optional] A Genius API token if you want to fetch lyrics from Genius
+- [Optional] Genius API token if you want to fetch lyrics from Genius
+- [Optional] Spotify cookie value if you want to fetch lyrics from Spotify
+- [Optional] OpenAI API token if you want to use LLM correction of the transcribed lyrics
+- [Optional] AudioShake API token if you want to use a much higher quality (but paid) API for lyrics transcription
 
 ```
 pip install lyrics-transcriber
@@ -26,6 +29,23 @@ pip install lyrics-transcriber
 > **Warning**
 > The package published to PyPI was created by manually editing `poetry.lock` to remove [triton](https://github.com/openai/triton), as it is technically a sub-dependency from openai-whisper but is currently only supported on Linux (whisper still works fine without it, and I want this package to be usable on any platform)
 
+## Docker
+
+You can use the pre-built container image `beveradb/lyrics-transcriber:0.16.0` on Docker hub if you want, here's an example:
+
+```sh
+docker run \
+ -v `pwd`/input:/input \
+ -v `pwd`/output:/output \
+beveradb/lyrics-transcriber:0.16.0 \
+ --log_level debug \
+ --output_dir /output \
+ --render_video \
+ --video_background_image /input/your-background-image.png \
+ --video_resolution 360p \
+ /input/song.flac
+```
+
 ## Usage 🚀
 
 ### As a standalone CLI

lyrics_transcriber-0.16.0/lyrics_transcriber/audioshake_transcriber.py

@@ -0,0 +1,35 @@
+import logging
+import requests
+
+
+class AudioShakeTranscriber:
+    def __init__(self, api_token, log_level=logging.DEBUG):
+        self.api_token = api_token
+        self.logger = logging.getLogger(__name__)
+        self.logger.setLevel(log_level)
+
+    def transcribe(self, audio_filepath):
+        # This is a placeholder for the actual AudioShake API implementation
+        self.logger.info(f"Transcribing {audio_filepath} using AudioShake API")
+
+        self.logger.debug(f"AudioShake API token: {self.api_token}")
+        # TODO: Implement the actual API call to AudioShake
+        # For now, we'll return a dummy result
+        return {
+            "transcription_data_dict": {
+                "segments": [
+                    {
+                        "start": 0,
+                        "end": 5,
+                        "text": "This is a dummy transcription",
+                        "words": [
+                            {"text": "This", "start": 0, "end": 1},
+                            {"text": "is", "start": 1, "end": 2},
+                            {"text": "a", "start": 2, "end": 3},
+                            {"text": "dummy", "start": 3, "end": 4},
+                            {"text": "transcription", "start": 4, "end": 5},
+                        ],
+                    }
+                ]
+            }
+        }

{lyrics_transcriber-0.14.0 → lyrics_transcriber-0.16.0}/lyrics_transcriber/transcriber.py

@@ -22,6 +22,8 @@ class LyricsTranscriber:
         audio_filepath,
         artist=None,
         title=None,
+        openai_api_key=None,
+        audioshake_api_token=None,
         genius_api_token=None,
         spotify_cookie=None,
         output_dir=None,
@@ -59,23 +61,30 @@ class LyricsTranscriber:
         self.title = title
         self.song_known = self.artist is not None and self.title is not None
 
+        self.openai_api_key = os.getenv("OPENAI_API_KEY", default=openai_api_key)
         self.genius_api_token = os.getenv("GENIUS_API_TOKEN", default=genius_api_token)
         self.spotify_cookie = os.getenv("SPOTIFY_COOKIE_SP_DC", default=spotify_cookie)
+        self.audioshake_api_token = os.getenv("AUDIOSHAKE_TOKEN", default=audioshake_api_token)
 
         self.transcription_model = transcription_model
         self.llm_model = llm_model
         self.llm_prompt_matching = llm_prompt_matching
         self.llm_prompt_correction = llm_prompt_correction
 
-        self.openai_client = OpenAI()
+        self.openai_client = None
 
-        # Uncomment for local models e.g. with ollama
-        # self.openai_client = OpenAI(
-        #     base_url="http://localhost:11434/v1",
-        #     api_key="ollama",
-        # )
+        if self.openai_api_key:
+            self.openai_client = OpenAI(api_key=self.openai_api_key)
 
-        self.openai_client.log = self.log_level
+            # Uncomment for local models e.g. with ollama
+            # self.openai_client = OpenAI(
+            #     base_url="http://localhost:11434/v1",
+            #     api_key="ollama",
+            # )
+
+            self.openai_client.log = self.log_level
+        else:
+            self.logger.error("No OpenAI API key found, no correction will be applied to transcription")
 
         self.render_video = render_video
         self.video_resolution = video_resolution
@@ -154,8 +163,13 @@ class LyricsTranscriber:
 
         self.validate_lyrics_match_song()
 
-        self.write_corrected_lyrics_data_file()
-        self.write_corrected_lyrics_plain_text()
+        if self.openai_client:
+            self.write_corrected_lyrics_data_file()
+            self.write_corrected_lyrics_plain_text()
+        else:
+            self.logger.warning("Skipping LLM correction as no OpenAI client is available")
+            self.outputs["corrected_lyrics_data_dict"] = self.outputs["transcription_data_dict"]
+            self.write_corrected_lyrics_plain_text()
 
         self.calculate_singing_percentage()
 
@@ -169,7 +183,8 @@ class LyricsTranscriber:
         self.copy_files_to_output_dir()
         self.calculate_llm_costs()
 
-        self.openai_client.close()
+        if self.openai_client:
+            self.openai_client.close()
 
         return self.outputs
 
@@ -198,41 +213,55 @@ class LyricsTranscriber:
             online_lyrics_text_key = f"{online_lyrics_source}_lyrics_text"
             online_lyrics_filepath_key = f"{online_lyrics_source}_lyrics_filepath"
 
-            if online_lyrics_text_key not in self.outputs:
+            if online_lyrics_text_key not in self.outputs or self.outputs[online_lyrics_text_key] is None:
                 continue
 
-            data_input_str = (
-                f'Data input 1:\n{self.outputs["transcribed_lyrics_text"]}\nData input 2:\n{self.outputs[online_lyrics_text_key]}\n'
-            )
-
-            # self.logger.debug(f"system_prompt:\n{system_prompt}\ndata_input_str:\n{data_input_str}")
-
-            self.logger.debug(f"making API call to LLM model {self.llm_model} to validate {online_lyrics_source} lyrics match")
-            response = self.openai_client.chat.completions.create(
-                model=self.llm_model,
-                messages=[{"role": "system", "content": llm_matching_instructions}, {"role": "user", "content": data_input_str}],
-            )
+            if self.openai_client:
+                data_input_str = (
+                    f'Data input 1:\n{self.outputs["transcribed_lyrics_text"]}\nData input 2:\n{self.outputs[online_lyrics_text_key]}\n'
+                )
 
-            message = response.choices[0].message.content
-            finish_reason = response.choices[0].finish_reason
+                self.logger.debug(f"making API call to LLM model {self.llm_model} to validate {online_lyrics_source} lyrics match")
+                response = self.openai_client.chat.completions.create(
+                    model=self.llm_model,
+                    messages=[{"role": "system", "content": llm_matching_instructions}, {"role": "user", "content": data_input_str}],
+                )
 
-            self.outputs["llm_token_usage"]["input"] += response.usage.prompt_tokens
-            self.outputs["llm_token_usage"]["output"] += response.usage.completion_tokens
+                message = response.choices[0].message.content
+                finish_reason = response.choices[0].finish_reason
 
-            # self.logger.debug(f"LLM API response finish_reason: {finish_reason} message: \n{message}")
+                self.outputs["llm_token_usage"]["input"] += response.usage.prompt_tokens
+                self.outputs["llm_token_usage"]["output"] += response.usage.completion_tokens
 
-            if finish_reason == "stop":
-                if message == "Yes":
-                    self.logger.info(f"{online_lyrics_source} lyrics successfully validated to match transcription")
+                if finish_reason == "stop":
+                    if message == "Yes":
+                        self.logger.info(f"{online_lyrics_source} lyrics successfully validated to match transcription")
+                        at_least_one_online_lyrics_validated = True
+                    elif message == "No":
+                        self.logger.warning(f"{online_lyrics_source} lyrics do not match transcription, deleting that source from outputs")
+                        self.outputs[online_lyrics_text_key] = None
+                        self.outputs[online_lyrics_filepath_key] = None
+                    else:
+                        self.logger.error(f"Unexpected response from LLM: {message}")
+                else:
+                    self.logger.warning(f"OpenAI API call did not finish successfully, finish_reason: {finish_reason}")
+            else:
+                # Fallback primitive word matching
+                self.logger.debug(f"Using primitive word matching to validate {online_lyrics_source} lyrics match")
+                transcribed_words = set(self.outputs["transcribed_lyrics_text"].split())
+                online_lyrics_words = set(self.outputs[online_lyrics_text_key].split())
+                common_words = transcribed_words & online_lyrics_words
+                match_percentage = len(common_words) / len(online_lyrics_words) * 100
+
+                if match_percentage >= 50:
+                    self.logger.info(
+                        f"{online_lyrics_source} lyrics successfully validated to match transcription with {match_percentage:.2f}% word match"
+                    )
                     at_least_one_online_lyrics_validated = True
-                elif message == "No":
+                else:
                     self.logger.warning(f"{online_lyrics_source} lyrics do not match transcription, deleting that source from outputs")
                     self.outputs[online_lyrics_text_key] = None
                     self.outputs[online_lyrics_filepath_key] = None
-                else:
-                    self.logger.error(f"Unexpected response from LLM: {message}")
-            else:
-                self.logger.warning(f"OpenAI API call did not finish successfully, finish_reason: {finish_reason}")
 
         self.logger.info(
             f"Completed validation of transcription using online lyrics sources. Match found: {at_least_one_online_lyrics_validated}"
@@ -242,9 +271,12 @@ class LyricsTranscriber:
             self.logger.error(
                 f"Lyrics from Genius and Spotify did not match the transcription. Please check artist and title are set correctly."
             )
-            raise Exception("Cannot proceed without internet lyrics to validate / correct transcription")
 
     def write_corrected_lyrics_data_file(self):
+        if not self.openai_client:
+            self.logger.warning("Skipping LLM correction as no OpenAI client is available")
+            return
+
         self.logger.debug("write_corrected_lyrics_data_file initiating OpenAI client")
 
         corrected_lyrics_data_json_cache_filepath = os.path.join(self.cache_dir, "lyrics-" + self.get_song_slug() + "-corrected.json")
@@ -583,51 +615,57 @@ class LyricsTranscriber:
                     f.write(line)
 
     def create_screens(self):
-        self.logger.debug(f"create_screens beginning generation of screens from whisper results")
+        self.logger.debug("create_screens beginning generation of screens from whisper results")
         screens: List[subtitles.LyricsScreen] = []
-        line: Optional[subtitles.LyricsLine] = None
         screen: Optional[subtitles.LyricsScreen] = None
 
-        lines_in_current_screen = 0
+        max_lines_per_screen = 4
+        max_line_length = 36  # Maximum characters per line
+        self.logger.debug(f"Max lines per screen: {max_lines_per_screen}, Max line length: {max_line_length}")
+
         for segment in self.outputs["corrected_lyrics_data_dict"]["segments"]:
-            self.logger.debug(f"lines_in_current_screen: {lines_in_current_screen} segment: {segment['text']}")
-            if screen is None:
-                self.logger.debug(f"screen is none, creating new LyricsScreen")
-                screen = subtitles.LyricsScreen()
-                screen.video_size = self.video_resolution_num
-                screen.line_height = self.line_height
-            if line is None:
-                self.logger.debug(f"line is none, creating new LyricsLine")
-                line = subtitles.LyricsLine()
-
-            num_words_in_segment = len(segment["words"])
-            for word_index, word in enumerate(segment["words"]):
-                segment = subtitles.LyricSegment(
+            self.logger.debug(f"Processing segment: {segment['text']}")
+            if screen is None or len(screen.lines) >= max_lines_per_screen:
+                screen = subtitles.LyricsScreen(video_size=self.video_resolution_num, line_height=self.line_height, logger=self.logger)
+                screens.append(screen)
+                self.logger.debug(f"Created new screen. Total screens: {len(screens)}")
+
+            words = segment["words"]
+            current_line = subtitles.LyricsLine()
+            current_line_text = ""
+            self.logger.debug(f"Processing {len(words)} words in segment")
+
+            for word in words:
+                self.logger.debug(f"Processing word: '{word['text']}'")
+                if len(current_line_text) + len(word["text"]) + 1 > max_line_length or (current_line_text and word["text"][0].isupper()):
+                    self.logger.debug(f"Current line would exceed max length or new capitalized word. Line: '{current_line_text}'")
+                    if current_line.segments:
+                        screen.lines.append(current_line)
+                        self.logger.debug(f"Added line to screen. Lines on current screen: {len(screen.lines)}")
+                        if len(screen.lines) >= max_lines_per_screen:
+                            screen = subtitles.LyricsScreen(
+                                video_size=self.video_resolution_num,
+                                line_height=self.line_height,
+                                logger=self.logger,
+                            )
+                            screens.append(screen)
+                            self.logger.debug(f"Screen full, created new screen. Total screens: {len(screens)}")
+                    current_line = subtitles.LyricsLine()
+                    current_line_text = ""
+                    self.logger.debug("Reset current line")
+
+                current_line_text += (" " if current_line_text else "") + word["text"]
+                lyric_segment = subtitles.LyricSegment(
                     text=word["text"], ts=timedelta(seconds=word["start"]), end_ts=timedelta(seconds=word["end"])
                 )
-                line.segments.append(segment)
-
-                # If word is last in the line, add line to screen and start new line
-                # Before looping to the next word
-                if word_index == num_words_in_segment - 1:
-                    self.logger.debug(f"word_index is last in segment, adding line to screen and starting new line")
-                    screen.lines.append(line)
-                    lines_in_current_screen += 1
-                    line = None
-
-            # If current screen has 2 lines already, add screen to list and start new screen
-            # Before looping to the next line
-            if lines_in_current_screen == 2:
-                self.logger.debug(f"lines_in_current_screen is 2, adding screen to list and starting new screen")
-                screens.append(screen)
-                screen = None
-                lines_in_current_screen = 0
+                current_line.segments.append(lyric_segment)
+                self.logger.debug(f"Added word to current line. Current line: '{current_line_text}'")
 
-        if line is not None:
-            screen.lines.append(line)  # type: ignore[union-attr]
-        if screen is not None and len(screen.lines) > 0:
-            screens.append(screen)  # type: ignore[arg-type]
+            if current_line.segments:
+                screen.lines.append(current_line)
+                self.logger.debug(f"Added final line of segment to screen. Lines on current screen: {len(screen.lines)}")
 
+        self.logger.debug(f"Finished creating screens. Total screens created: {len(screens)}")
         return screens
 
     def write_ass_file(self):
@@ -760,7 +798,10 @@ class LyricsTranscriber:
 
     def write_transcribed_lyrics_plain_text(self):
         if self.outputs["transcription_data_dict"]:
-            transcribed_lyrics_text_filepath = os.path.join(self.cache_dir, "lyrics-" + self.get_song_slug() + "-transcribed.txt")
+            transcription_cache_suffix = "-audioshake-transcribed.txt" if self.audioshake_api_token else "-whisper-transcribed.txt"
+            self.logger.debug(f"transcription_cache_suffix: {transcription_cache_suffix}")
+
+            transcribed_lyrics_text_filepath = os.path.join(self.cache_dir, "lyrics-" + self.get_song_slug() + transcription_cache_suffix)
             self.outputs["transcribed_lyrics_text_filepath"] = transcribed_lyrics_text_filepath
 
             self.outputs["transcribed_lyrics_text"] = ""
@@ -773,8 +814,109 @@ class LyricsTranscriber:
         else:
             raise Exception("Cannot write transcribed lyrics plain text as transcription_data_dict is not set")
 
+    def find_best_split_point(self, text, max_length):
+        self.logger.debug(f"Finding best split point for text: '{text}' (max_length: {max_length})")
+        words = text.split()
+        mid_word_index = len(words) // 2
+        mid_point = len(" ".join(words[:mid_word_index]))
+        self.logger.debug(f"Mid point is at character {mid_point}")
+
+        # Check for a comma within one or two words of the middle word
+        if "," in text:
+            comma_indices = [i for i, char in enumerate(text) if char == ","]
+            self.logger.debug(f"Found commas at indices: {comma_indices}")
+            for index in comma_indices:
+                if abs(mid_point - index) < 20 and len(text[: index + 1].strip()) <= max_length:
+                    self.logger.debug(f"Choosing comma at index {index} as split point")
+                    return index + 1  # Include the comma in the first part
+
+        # Check for 'and'
+        if " and " in text:
+            and_indices = [m.start() for m in re.finditer(" and ", text)]
+            self.logger.debug(f"Found 'and' at indices: {and_indices}")
+            for index in sorted(and_indices, key=lambda x: abs(x - mid_point)):
+                if len(text[: index + len(" and ")].strip()) <= max_length:
+                    self.logger.debug(f"Choosing 'and' at index {index} as split point")
+                    return index + len(" and ")
+
+        # Check for words starting with a capital letter
+        capital_word_indices = [m.start() for m in re.finditer(r"\s[A-Z]", text)]
+        self.logger.debug(f"Found capital words at indices: {capital_word_indices}")
+        for index in sorted(capital_word_indices, key=lambda x: abs(x - mid_point)):
+            if index > 0 and len(text[:index].strip()) <= max_length:
+                self.logger.debug(f"Choosing capital word at index {index} as split point")
+                return index
+
+        # If no better split point is found, try splitting at the middle word
+        if len(words) > 2 and mid_word_index > 0:
+            split_at_middle = len(" ".join(words[:mid_word_index]))
+            if split_at_middle <= max_length:
+                self.logger.debug(f"Choosing middle word split at index {split_at_middle}")
+                return split_at_middle
+
+        # If the text is still too long, forcibly split at the maximum length
+        self.logger.debug(f"No suitable split point found, forcibly splitting at max_length {max_length}")
+        return max_length
+
+    def split_long_segments(self, segments, max_length):
+        self.logger.debug(f"Splitting long segments (max_length: {max_length})")
+        new_segments = []
+        for segment in segments:
+            text = segment["text"]
+            self.logger.debug(f"Processing segment: '{text}' (length: {len(text)})")
+            if len(text) <= max_length:
+                self.logger.debug("Segment is within max_length, keeping as is")
+                new_segments.append(segment)
+            else:
+                self.logger.debug("Segment exceeds max_length, splitting")
+                meta_words = segment["words"]
+                current_text = ""
+                current_start = segment["start"]
+                current_words = []
+
+                for i, meta in enumerate(meta_words):
+                    word = meta["text"]
+                    if current_text:
+                        current_text += " "
+                    current_text += word
+                    current_words.append(meta)
+
+                    should_split = len(current_text) > max_length or (i > 0 and word[0].isupper())
+                    if should_split:
+                        self.logger.debug(f"Splitting at: '{current_text}'")
+                        # If splitting due to capitalization, don't include the capitalized word
+                        if word[0].isupper() and len(current_text.strip()) > len(word):
+                            split_text = current_text[: -(len(word) + 1)].strip()
+                            current_words = current_words[:-1]
+                        else:
+                            split_text = current_text.strip()
+
+                        new_segment = {"text": split_text, "start": current_start, "end": current_words[-1]["end"], "words": current_words}
+                        new_segments.append(new_segment)
+                        self.logger.debug(f"Added new segment: {new_segment}")
+
+                        # Reset for next segment
+                        if word[0].isupper() and len(current_text.strip()) > len(word):
+                            current_text = word
+                            current_words = [meta]
+                        else:
+                            current_text = ""
+                            current_words = []
+                        current_start = meta["start"]
+
+                # Add any remaining text as a final segment
+                if current_text:
+                    self.logger.debug(f"Adding final segment: '{current_text}'")
+                    new_segments.append(
+                        {"text": current_text.strip(), "start": current_start, "end": segment["end"], "words": current_words}
+                    )
+
+        self.logger.debug(f"Splitting complete. Original segments: {len(segments)}, New segments: {len(new_segments)}")
+        return new_segments
+
     def transcribe(self):
-        self.outputs["transcription_data_filepath"] = self.get_cache_filepath(".json")
+        transcription_cache_suffix = "-audioshake" if self.audioshake_api_token else "-whisper"
+        self.outputs["transcription_data_filepath"] = self.get_cache_filepath(f"{transcription_cache_suffix}.json")
 
         whisper_cache_filepath = self.outputs["transcription_data_filepath"]
         if os.path.isfile(whisper_cache_filepath):
@@ -783,15 +925,26 @@ class LyricsTranscriber:
                 self.outputs["transcription_data_dict"] = json.load(cache_file)
                 return
 
-        self.logger.debug(f"no cached transcription file found, running whisper transcribe with model: {self.transcription_model}")
-        audio = whisper.load_audio(self.audio_filepath)
-        model = whisper.load_model(self.transcription_model, device="cpu")
-        result = whisper.transcribe(model, audio, language="en")
+        if self.audioshake_api_token:
+            self.logger.debug(f"Using AudioShake API for transcription")
+            from .audioshake_transcriber import AudioShakeTranscriber
 
-        self.logger.debug(f"transcription complete, performing post-processing cleanup")
-
-        # Remove segments with no words, only music
-        result["segments"] = [segment for segment in result["segments"] if segment["text"].strip() != "Music"]
+            audioshake = AudioShakeTranscriber(self.audioshake_api_token, log_level=self.log_level)
+            result = audioshake.transcribe(self.audio_filepath)
+        else:
+            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")
+            result = whisper.transcribe(model, audio, language="en", vad="auditok", beam_size=5, temperature=0.2, best_of=5)
+
+            # Remove segments with no words, only music
+            result["segments"] = [segment for segment in result["segments"] if segment["text"].strip() != "Music"]
+            self.logger.debug(f"Removed 'Music' segments. Remaining segments: {len(result['segments'])}")
+
+            # Split long segments
+            self.logger.debug("Starting to split long segments")
+            result["segments"] = self.split_long_segments(result["segments"], max_length=36)
+            self.logger.debug(f"Finished splitting segments. Total segments after splitting: {len(result['segments'])}")
 
         self.logger.debug(f"writing transcription data JSON to cache file: {whisper_cache_filepath}")
         with open(whisper_cache_filepath, "w") as cache_file:

{lyrics_transcriber-0.14.0 → lyrics_transcriber-0.16.0}/lyrics_transcriber/utils/cli.py

@@ -34,6 +34,11 @@ def main():
         default=None,
         help="Optional: song title for lyrics lookup and auto-correction",
     )
+    parser.add_argument(
+        "--audioshake_api_token",
+        default=None,
+        help="Optional: AudioShake API token for lyrics transcription and alignment. Can also be set with AUDIOSHAKE_API_TOKEN env var.",
+    )
     parser.add_argument(
         "--genius_api_token",
         default=None,
@@ -77,7 +82,7 @@ def main():
 
     parser.add_argument(
         "--video_resolution",
-        default="4k",
+        default="360p",
         help="Optional: resolution of the karaoke video to render. Must be one of: 4k, 1080p, 720p, 360p. Default: 360p",
     )
 
@@ -93,6 +98,12 @@ def main():
         help="Optional: color to use for karaoke video background, in hex format or FFmpeg color name. Default: black",
     )
 
+    parser.add_argument(
+        "--openai_api_key",
+        default=None,
+        help="Optional: OpenAI API key for LLM model usage. Can also be set with OPENAI_API_KEY env var.",
+    )
+
     args = parser.parse_args()
 
     log_level = getattr(logging, args.log_level.upper())
@@ -114,8 +125,10 @@ def main():
 
     transcriber = LyricsTranscriber(
         args.audio_filepath,
+        audioshake_api_token=args.audioshake_api_token,
         genius_api_token=args.genius_api_token,
         spotify_cookie=args.spotify_cookie,
+        openai_api_key=args.openai_api_key,
         artist=args.artist,
         title=args.title,
         output_dir=args.output_dir,

{lyrics_transcriber-0.14.0 → lyrics_transcriber-0.16.0}/lyrics_transcriber/utils/subtitles.py

@@ -5,6 +5,7 @@ import json
 import itertools
 from pathlib import Path
 from enum import IntEnum
+import logging
 
 from . import ass
 
@@ -85,21 +86,19 @@ class LyricsLine:
     def __str__(self):
         return "".join([f"{{{s.text}}}" for s in self.segments])
 
-    def as_ass_event(
-        self,
-        screen_start: timedelta,
-        screen_end: timedelta,
-        style: ass.ASS.Style,
-        top_margin: int,
-    ):
+    def as_ass_event(self, screen_start: timedelta, screen_end: timedelta, style: ass.ASS.Style, y_position: int):
         e = ass.ASS.Event()
         e.type = "Dialogue"
         e.Layer = 0
         e.Style = style
         e.Start = screen_start.total_seconds()
         e.End = screen_end.total_seconds()
-        e.MarginV = top_margin
+        e.MarginV = y_position
         e.Text = self.decorate_ass_line(self.segments, screen_start)
+
+        # Set alignment to top-center
+        e.Text = "{\\an8}" + e.Text
+
         return e
 
     def decorate_ass_line(self, segments, screen_start_ts: timedelta):
@@ -137,6 +136,7 @@ class LyricsScreen:
     start_ts: Optional[timedelta] = None
     video_size: Tuple[int, int] = None
     line_height: int = None
+    logger: logging.Logger = None
 
     @property
     def end_ts(self) -> timedelta:
@@ -145,10 +145,36 @@ class LyricsScreen:
     def get_line_y(self, line_num: int) -> int:
         _, h = self.video_size
         line_count = len(self.lines)
-        return (h / 2) - (line_count * self.line_height / 2) + (line_num * self.line_height)
+        total_height = line_count * self.line_height
+
+        # Calculate the top margin to center the lyrics block
+        top_margin = (h - total_height) / 2
+
+        # Calculate the y-position for this specific line
+        line_y = top_margin + (line_num * self.line_height)
+
+        # if self.logger:
+        #     self.logger.debug(f"Line {line_num + 1} positioning:")
+        #     self.logger.debug(f"  Video height: {h}")
+        #     self.logger.debug(f"  Total lines: {line_count}")
+        #     self.logger.debug(f"  Line height: {self.line_height}")
+        #     self.logger.debug(f"  Total lyrics height: {total_height}")
+        #     self.logger.debug(f"  Top margin: {top_margin}")
+        #     self.logger.debug(f"  Line y: {line_y}")
+
+        return int(line_y)
 
     def as_ass_events(self, style: ass.ASS.Style) -> List[ass.ASS.Event]:
-        return [line.as_ass_event(self.start_ts, self.end_ts, style, self.get_line_y(i)) for i, line in enumerate(self.lines)]
+        events = []
+        for i, line in enumerate(self.lines):
+            y_position = self.get_line_y(i)
+            
+            # if self.logger:
+            #     self.logger.debug(f"Creating ASS event for line {i + 1} at y-position: {y_position}")
+            
+            event = line.as_ass_event(self.start_ts, self.end_ts, style, y_position)
+            events.append(event)
+        return events
 
     def __str__(self):
         lines = [f"{self.start_ts} - {self.end_ts}:"]
@@ -264,7 +290,7 @@ def create_styled_subtitles(
     style.BorderStyle = 1
     style.Outline = 1
     style.Shadow = 0
-    style.Alignment = ass.ASS.ALIGN_MIDDLE_CENTER
+    style.Alignment = ass.ASS.ALIGN_TOP_CENTER
     style.MarginL = 0
     style.MarginR = 0
     style.MarginV = 0

{lyrics_transcriber-0.14.0 → lyrics_transcriber-0.16.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "lyrics-transcriber"
-version = "0.14.0"
+version = "0.16.0"
 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"
@@ -28,6 +28,7 @@ syrics = ">=0"
 openai = "^1"
 openai-whisper = ">=20231117"
 transformers = ">=4"
+auditok = ">=0.2"
 whisper-timestamped = ">=1"
 # 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.