Mirroristas/whisper
2
0
Fork 0
mirror of https://github.com/openai/whisper.git synced 2025-11-24 06:26:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add option to carry initial_prompt with the sliding window

Browse Source 
Add an option `carry_initial_prompt = False` to `whisper.transcribe()`.
When set to `True`, `initial_prompt` is prepended to each internal `decode()` call's `prompt`.
If there is not enough context space at the start of the prompt, the prompt is left-sliced to make space.
This commit is contained in:
Kittsil 2024-09-18 22:03:25 -05:00
parent 32d55d5d76
commit ce140e4f7a
1 changed files with 17 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
whisper/transcribe.py
View File

@ -46,6 +46,7 @@ def transcribe(
no_speech_threshold: Optional[float] = 0.6, no_speech_threshold: Optional[float] = 0.6,
condition_on_previous_text: bool = True, condition_on_previous_text: bool = True,
initial_prompt: Optional[str] = None, initial_prompt: Optional[str] = None,
carry_initial_prompt: bool = False,
word_timestamps: bool = False, word_timestamps: bool = False,
prepend_punctuations: str = "\"'“¿([{-", prepend_punctuations: str = "\"'“¿([{-",
append_punctuations: str = "\"'.。,，!！?？:：”)]}、", append_punctuations: str = "\"'.。,，!！?？:：”)]}、",
@ -102,6 +103,11 @@ def transcribe(
"prompt-engineer" a context for transcription, e.g. custom vocabularies or proper nouns "prompt-engineer" a context for transcription, e.g. custom vocabularies or proper nouns
to make it more likely to predict those word correctly. to make it more likely to predict those word correctly.
carry_initial_prompt: bool
If carry_initial_prompt is True, `initial_prompt` is prepended to the prompt of each internal
`decode()` call. If there is not enough context space at the start of the prompt, it is
left-sliced to make space.
decode_options: dict decode_options: dict
Keyword arguments to construct `DecodingOptions` instances Keyword arguments to construct `DecodingOptions` instances
@ -227,9 +233,11 @@ def transcribe(
all_segments = [] all_segments = []
prompt_reset_since = 0 prompt_reset_since = 0
remaining_prompt_length = model.dims.n_text_ctx // 2 - 1
if initial_prompt is not None: if initial_prompt is not None:
initial_prompt_tokens = tokenizer.encode(" " + initial_prompt.strip()) initial_prompt_tokens = tokenizer.encode(" " + initial_prompt.strip())
all_tokens.extend(initial_prompt_tokens) all_tokens.extend(initial_prompt_tokens)
remaining_prompt_length -= len(initial_prompt_tokens)
else: else:
initial_prompt_tokens = [] initial_prompt_tokens = []
@ -275,7 +283,13 @@ def transcribe(
segment_duration = segment_size * HOP_LENGTH / SAMPLE_RATE segment_duration = segment_size * HOP_LENGTH / SAMPLE_RATE
mel_segment = pad_or_trim(mel_segment, N_FRAMES).to(model.device).to(dtype) mel_segment = pad_or_trim(mel_segment, N_FRAMES).to(model.device).to(dtype)
decode_options["prompt"] = all_tokens[prompt_reset_since:] prompt_reseed = all_tokens[prompt_reset_since:]
if carry_initial_prompt:
prompt_reseed = (
initial_prompt_tokens + prompt_reseed[-remaining_prompt_length:]
)
decode_options["prompt"] = prompt_reseed
result: DecodingResult = decode_with_fallback(mel_segment) result: DecodingResult = decode_with_fallback(mel_segment)
tokens = torch.tensor(result.tokens) tokens = torch.tensor(result.tokens)
@ -529,6 +543,8 @@ def cli():
parser.add_argument("--suppress_tokens", type=str, default="-1", help="comma-separated list of token ids to suppress during sampling; '-1' will suppress most special characters except common punctuations") parser.add_argument("--suppress_tokens", type=str, default="-1", help="comma-separated list of token ids to suppress during sampling; '-1' will suppress most special characters except common punctuations")
parser.add_argument("--initial_prompt", type=str, default=None, help="optional text to provide as a prompt for the first window.") parser.add_argument("--initial_prompt", type=str, default=None, help="optional text to provide as a prompt for the first window.")
parser.add_argument("--carry_initial_prompt", type=str2bool, default=False, help="if True, prepend initial_prompt to every internal decode() call. May reduce the effectiveness of condition_on_previous_text")
parser.add_argument("--condition_on_previous_text", type=str2bool, default=True, help="if True, provide the previous output of the model as a prompt for the next window; disabling may make the text inconsistent across windows, but the model becomes less prone to getting stuck in a failure loop") parser.add_argument("--condition_on_previous_text", type=str2bool, default=True, help="if True, provide the previous output of the model as a prompt for the next window; disabling may make the text inconsistent across windows, but the model becomes less prone to getting stuck in a failure loop")
parser.add_argument("--fp16", type=str2bool, default=True, help="whether to perform inference in fp16; True by default") parser.add_argument("--fp16", type=str2bool, default=True, help="whether to perform inference in fp16; True by default")