o 8wi@,@s ddlmZGdddeZdS))SpeakerDiarizationProtocolc@seZdZdZddZddZddZdd Zd d Zd d Z ddZ ddZ ddZ ddZ ddZddZddZddZddZd S)!SpeakerSpottingProtocolaSpeaker spotting protocol Parameters ---------- preprocessors : dict or (key, preprocessor) iterable When provided, each protocol item (dictionary) are preprocessed, such that item[key] = preprocessor(item). In case 'preprocessor' is not callable, it should be a string containing placeholder for item keys (e.g. {'audio': '/path/to/{uri}.wav'}) cCtd)Nz=Custom speaker spotting protocol should implement "trn_iter".NotImplementedErrorselfr h/home/ubuntu/sommelier/.venv/lib/python3.10/site-packages/pyannote/database/protocol/speaker_spotting.pytrn_iter-z SpeakerSpottingProtocol.trn_itercCdSNr rr r r trn_enrol_iter2z&SpeakerSpottingProtocol.trn_enrol_itercCr rr rr r r trn_try_iter5rz$SpeakerSpottingProtocol.trn_try_itercCr)Nz=Custom speaker spotting protocol should implement "dev_iter".rrr r r dev_iter8r z SpeakerSpottingProtocol.dev_itercCr)NzCCustom speaker spotting protocol should implement "dev_enrol_iter".rrr r r dev_enrol_iter=r z&SpeakerSpottingProtocol.dev_enrol_itercCr)NzACustom speaker spotting protocol should implement "dev_try_iter".rrr r r dev_try_iterBr z$SpeakerSpottingProtocol.dev_try_itercCr)Nz=Custom speaker spotting protocol should implement "tst_iter".rrr r r tst_iterGr z SpeakerSpottingProtocol.tst_itercCr)NzCCustom speaker spotting protocol should implement "tst_enrol_iter".rrr r r tst_enrol_iterLr z&SpeakerSpottingProtocol.tst_enrol_itercCr)NzACustom speaker spotting protocol should implement "tst_try_iter".rrr r r tst_try_iterQr z$SpeakerSpottingProtocol.tst_try_itercc$|}|D]}||VqdS)a;Iterate over the enrolments of the train set Yields dictionaries with the followings keys: * uri: str unique audio file identifier * database: str unique database identifier * model_id: str unique model identifier (the same speaker might have different models) * enrol_with: pyannote.core.Timeline parts of the audio file to use for enrolment as well as keys coming from the provided preprocessors (e.g. 'audio') Usage ----- >>> models = {} >>> for enrolment in protocol.train_enrolment(): ... # obtain path to audio file ... audio = enrolment['audio'] ... # obtain parts of the audio file to use for enrolment ... enrol_with = enrolment['enrol_with'] ... # this is where enrolment actually happens ... model = do_something(audio, enrol_with) ... # store models for later use ... model_id = enrolment['model_id'] ... models[model_id] = model N)r preprocessr generatorcurrent_enrolmentr r r train_enrolmentV  z'SpeakerSpottingProtocol.train_enrolmentccr)aQIterate over the trials of the train set Yields dictionaries with the followings keys: * uri: str unique audio file identifier * database: str unique database identifier * try_with: pyannote.core.Segment, optional parts of the audio file where to look for the target speaker. default is to use the whole audio file * model_id: str unique identifier of the target * reference: pyannote.core.Timeline parts of the audio file where the target actually speaks. it might be empty in case of impostor trials. in case of genuine trials, it should be contained in `try_with` as well as keys coming from the provided preprocessors (e.g. 'audio') Usage ----- >>> for trial in protocol.train_trial(): ... # obtain path to audio file ... audio = trial['audio'] ... # obtain parts of the audio file to use for trial ... try_with = trial['try_with'] ... # this is where the trial actually happens ... model_id = trial['model_id'] ... score = do_something(audio, try_with, model_id) ... # optionally perform evaluation ... reference = trial['reference'] ... metric(reference, score) N)rrrr current_trialr r r train_trial{ %z#SpeakerSpottingProtocol.train_trialccr)aGIterate over the enrolments of the development set Yields dictionaries with the followings keys: * uri: str unique audio file identifier * database: str unique database identifier * model_id: str unique model identifier (the same speaker might have different models) * enrol_with: pyannote.core.Timeline parts of the audio file to use for enrolment as well as keys coming from the provided preprocessors (e.g. 'audio') Usage ----- >>> models = {} >>> for enrolment in protocol.development_enrolment(): ... # obtain path to audio file ... audio = enrolment['audio'] ... # obtain parts of the audio file to use for enrolment ... enrol_with = enrolment['enrol_with'] ... # this is where enrolment actually happens ... model = do_something(audio, enrol_with) ... # store models for later use ... model_id = enrolment['model_id'] ... models[model_id] = model N)rrrr r r development_enrolmentrz-SpeakerSpottingProtocol.development_enrolmentccr)a]Iterate over the trials of the development set Yields dictionaries with the followings keys: * uri: str unique audio file identifier * database: str unique database identifier * try_with: pyannote.core.Segment, optional parts of the audio file where to look for the target speaker. default is to use the whole audio file * model_id: str unique identifier of the target * reference: pyannote.core.Timeline parts of the audio file where the target actually speaks. it might be empty in case of impostor trials. in case of genuine trials, it should be contained in `try_with` as well as keys coming from the provided preprocessors (e.g. 'audio') Usage ----- >>> for trial in protocol.development_trial(): ... # obtain path to audio file ... audio = trial['audio'] ... # obtain parts of the audio file to use for trial ... try_with = trial['try_with'] ... # this is where the trial actually happens ... model_id = trial['model_id'] ... score = do_something(audio, try_with, model_id) ... # optionally perform evaluation ... reference = trial['reference'] ... metric(reference, score) N)rrrr r r development_trialr"z)SpeakerSpottingProtocol.development_trialccr)a9Iterate over the enrolments of the test set Yields dictionaries with the followings keys: * uri: str unique audio file identifier * database: str unique database identifier * model_id: str unique model identifier (the same speaker might have different models) * enrol_with: pyannote.core.Timeline parts of the audio file to use for enrolment as well as keys coming from the provided preprocessors (e.g. 'audio') Usage ----- >>> models = {} >>> for enrolment in protocol.test_enrolment(): ... # obtain path to audio file ... audio = enrolment['audio'] ... # obtain parts of the audio file to use for enrolment ... enrol_with = enrolment['enrol_with'] ... # this is where enrolment actually happens ... model = do_something(audio, enrol_with) ... # store models for later use ... model_id = enrolment['model_id'] ... models[model_id] = model N)rrrr r r test_enrolmentrz&SpeakerSpottingProtocol.test_enrolmentccr)aOIterate over the trials of the test set Yields dictionaries with the followings keys: * uri: str unique audio file identifier * database: str unique database identifier * try_with: pyannote.core.Segment, optional parts of the audio file where to look for the target speaker. default is to use the whole audio file * model_id: str unique identifier of the target * reference: pyannote.core.Timeline parts of the audio file where the target actually speaks. it might be empty in case of impostor trials. in case of genuine trials, it should be contained in `try_with` as well as keys coming from the provided preprocessors (e.g. 'audio') Usage ----- >>> for trial in protocol.test_trial(): ... # obtain path to audio file ... audio = trial['audio'] ... # obtain parts of the audio file to use for trial ... try_with = trial['try_with'] ... # this is where the trial actually happens ... model_id = trial['model_id'] ... score = do_something(audio, try_with, model_id) ... # optionally perform evaluation ... reference = trial['reference'] ... metric(reference, score) N)rrrr r r test_trialr"z"SpeakerSpottingProtocol.test_trialN)__name__ __module__ __qualname____doc__r rrrrrrrrrr!r#r$r%r&r r r r r!s" %*%* %rN)speaker_diarizationrrr r r r s