WIP: Vcc2018 #611
WIP: Vcc2018 #611
Conversation
SupervisionSet contains 4 supervisions that start at 0 for recording XY for all recordings because each recording has four MOS supervisions. Plus there are 291 recordings which do not have suppervisions - TODO investigate
| @@ -56,3 +56,4 @@ | |||
| from .voxceleb import * | |||
| from .wenet_speech import * | |||
| from .yesno import * | |||
| from .vcc2018 import * | |||
There was a problem hiding this comment.
We try to keep the imports sorted in lexicographical order. Could you put this import above vctk?
| mos_scores = mos_dir / "vcc2018_evaluation_mos.txt" | ||
| assert mos_scores.is_file() | ||
| sim_scores = mos_dir / "vcc2018_evaluation_sim.txt" | ||
| sim_scores.is_file() |
There was a problem hiding this comment.
should this be an assert?
| f"Collecting reference target recordings for the VCC2018 challange from {reference_speech_dir}" | ||
| ) | ||
|
|
||
| # TODO |
| return {"recordings": recordings, "supervisions": supervisions} | ||
|
|
||
|
|
||
| def prepare_mos_supervisions( |
There was a problem hiding this comment.
If this function is only intended to be used inside this module, it is recommended to add a single underscore, i.e., _prepare_mos_supervisions(). Note that this does not enforce privacy but only indicates that this method should not be called directly.
| return SupervisionSet.from_segments(supervisions) | ||
|
|
||
|
|
||
| def load_vcc_results(path: Pathlike): |
There was a problem hiding this comment.
Same _load_vcc_results()
| """ | ||
| Returns pandas.DataFrame | ||
| """ | ||
| # """ |
There was a problem hiding this comment.
Use either multi-line comment (""" """) or single-line (#)
| recording_ids = set(mos["left_audio"].tolist()) | ||
| supervisions = [] | ||
| for recording_id_wav in tqdm(recording_ids, desc="Supervision creation"): | ||
| recording_id = recording_id_wav.rstrip(".wav") |
There was a problem hiding this comment.
If you make the recording_ids above as a set of Path types, then you could use .stem here instead.
There was a problem hiding this comment.
I know… but it is less readible with Path(recording_id_wav).stem since I literally need to use both recording_id wit hand without “.wav” for the original data and Lhotse use.
There was a problem hiding this comment.
recording_id_wav.name would give you the file name with the extension.
| def prepare_mos_supervisions( | ||
| mos_results_path, recordings: RecordingSet, id2trn: Dict[str, str] | ||
| ) -> SupervisionSet: | ||
| # TODO very slow -> make it faster it takes ~8min 170it/s |
There was a problem hiding this comment.
This may be because you are manipulating a Pandas dataframe. Pandas does a lot of book-keeping under the hood which makes it good for complicated operations, but here I think it may be better to just use something like a list of namedtuples, and then use groupby to group them.
|
@desh2608 I forgot about this WIP PR. I will address your comments next week. |
No description provided.