lyrics-transcriber 0.13.1__tar.gz → 0.15.0__tar.gz

{lyrics_transcriber-0.13.1 → lyrics_transcriber-0.15.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: lyrics-transcriber
-Version: 0.13.1
+Version: 0.15.0
 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)

{lyrics_transcriber-0.13.1 → lyrics_transcriber-0.15.0}/lyrics_transcriber/transcriber.py

@@ -29,7 +29,7 @@ class LyricsTranscriber:
         log_level=logging.DEBUG,
         log_formatter=None,
         transcription_model="medium",
-        llm_model="gpt-4-1106-preview",
+        llm_model="gpt-4o",
         llm_prompt_matching="lyrics_transcriber/llm_prompts/llm_prompt_lyrics_matching_andrew_handwritten_20231118.txt",
         llm_prompt_correction="lyrics_transcriber/llm_prompts/llm_prompt_lyrics_correction_andrew_handwritten_20231118.txt",
         render_video=False,
@@ -66,7 +66,15 @@ class LyricsTranscriber:
         self.llm_model = llm_model
         self.llm_prompt_matching = llm_prompt_matching
         self.llm_prompt_correction = llm_prompt_correction
+
         self.openai_client = OpenAI()
+
+        # 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
 
         self.render_video = render_video
@@ -391,8 +399,11 @@ class LyricsTranscriber:
             },
         }
 
-        input_cost = price_dollars_per_1000_tokens[self.llm_model]["input"] * (self.outputs["llm_token_usage"]["input"] / 1000)
-        output_cost = price_dollars_per_1000_tokens[self.llm_model]["output"] * (self.outputs["llm_token_usage"]["output"] / 1000)
+        input_price = price_dollars_per_1000_tokens.get(self.llm_model, {"input": 0, "output": 0})["input"]
+        output_price = price_dollars_per_1000_tokens.get(self.llm_model, {"input": 0, "output": 0})["output"]
+
+        input_cost = input_price * (self.outputs["llm_token_usage"]["input"] / 1000)
+        output_cost = output_price * (self.outputs["llm_token_usage"]["output"] / 1000)
 
         self.outputs["llm_costs_usd"]["input"] = round(input_cost, 3)
         self.outputs["llm_costs_usd"]["output"] = round(output_cost, 3)
@@ -775,7 +786,7 @@ class LyricsTranscriber:
         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")
+        result = whisper.transcribe(model, audio, language="en", vad="auditok", beam_size=5, temperature=0.2, best_of=5)
 
         self.logger.debug(f"transcription complete, performing post-processing cleanup")
 

{lyrics_transcriber-0.13.1 → lyrics_transcriber-0.15.0}/lyrics_transcriber/utils/cli.py

@@ -65,8 +65,8 @@ def main():
 
     parser.add_argument(
         "--llm_model",
-        default="gpt-4-1106-preview",
-        help="Optional: LLM model to use (currently only supports OpenAI chat completion models, e.g. gpt-4-1106-preview). Default: gpt-3.5-turbo-1106",
+        default="gpt-4o",
+        help="Optional: LLM model to use (currently only supports OpenAI chat completion compatible models",
     )
 
     parser.add_argument(

{lyrics_transcriber-0.13.1 → lyrics_transcriber-0.15.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "lyrics-transcriber"
-version = "0.13.1"
+version = "0.15.0"
 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.