data_example.md

Recommended

[NOTE] Single speaker works the same as the legacy version

#multi speaker (lab + wav)
your_zip.zip:
    |
    |
    lang_config.yaml
    |
    your_speaker_folder_1.lang1:
        |
        |
        data_1.wav
        data_1.lab
        .
        data_2.wav
        data_2.lab
        .
        data_3.wav
        data_3.lab
        .
        ...
    |
    your_speaker_folder_2.lang2:
        |
        |
        data_1.wav
        data_1.lab
        .
        data_2.wav
        data_2.lab
        .
        data_3.wav
        data_3.lab
        .[Uploading lang_config.yaml…]()

        ...

lang_config.yaml content:

languages: [] # to specify language dict according to the speaker folder's extension
extra_phonemes: [AP, SP] # any extras that are in your labels, eg. vf, q, cl
merged_phoneme_groups: [] # define phonemes group that has the same sound (like-phoneme) throughout the dataset across labeling systems

Example:

diffsinger_training_data.zip:
    |
    |
    lang_config.yaml
    |
    PJS.ja:
        (data)
    |
    Opencpop.zh:
        (data)

lang_config.yaml content:

languages: [ja, zh]
extra_phonemes: [AP, SP]
merged_phoneme_groups:
  - [zh/i, ja/i]
  - [zh/s, ja/s]

---

Legacy section (still works as single-dict)

[NOTE] .ds training has the same zip organization as lab + wav, but with only .ds files- no wav needed
#single speaker (lab + wav)
your_zip.zip:
    |
    |
    your_speaker_folder:
        |
        |
        data_1.wav
        data_1.lab
        .
        data_2.wav
        data_2.lab
        .
        data_3.wav
        data_3.lab
        .
        ...

#single speaker (csv + wav)
your_zip.zip:
    |
    |
    your_speaker_folder:
        |
        |
        wavs (folder named "wavs" containing all the wavs)
        .
        transcriptions.csv

#multi speaker (lab + wav)
your_zip.zip:
    |
    |
    your_speaker_folder_1:
        |
        |
        data_1.wav
        data_1.lab
        .
        data_2.wav
        data_2.lab
        .
        data_3.wav
        data_3.lab
        .
        ...
    your_speaker_folder_2:
        |
        |
        data_1.wav
        data_1.lab
        .
        data_2.wav
        data_2.lab
        .
        data_3.wav
        data_3.lab
        .
        ...

#multi speaker (csv + wav)
your_zip.zip:
    |
    |
    your_speaker_folder_1:
        |
        |
        wavs (folder named "wavs" containing all the wavs)
        .
        transcriptions.csv
    your_speaker_folder_2:
        |
        |
        wavs (folder named "wavs" containing all the wavs)
        .
        transcriptions.csv