IndexError: list index out of range

[omkarade]

[Konstant1982]

I have the same error, did you find a solution?

[mepc36]

Same error, any solution?

[mepc36]

@Konstant1982 Thanks to the Tacotron forums here I was able to get past this error by removing the final blank line in my metadata.csv file. Weird, because the train command still worked when I ran it with LJSpeech's own metadata.csv file, which has its own empty blank last line too...weird!

[gnmarten]

in my case, this exact error was due to an inconsistency in the metadata.csv file (rows having fewer than 2 columns).

You can run a sanity check like this to detect the faulty rows (and correct them manually)

filename = "metadata.csv"
with open(filename, "r", encoding="utf-8") as f:
    row_num = 0
    for line in f:
        cols = line.strip().split("|")
        if len(cols) < 2:
            print(f"Row {row_num + 1} has fewer than 2 columns: {line.strip()}")
        row_num += 1

[VArdulov]

So digging into the relevant code, it seems like the issue is coming from the format of the metadata file and the formatter that you choose. It looks like you're using the ljspeech formatter which I think is set as the default in many of the recipes for training/tuning. And if you are like me and you followed the guide to preparing the dataset in the documentation for TTS training/tuning, then you'll notice that either there is an error in the ljspeech formatter code that grabs cols[2] instead of cols[1] which is throwing the Index Out of Bounds error or the recommended format in the data preparation is not the LJSpeech format. Looking elsewhere in the guide it looks like the mozilla formatter does the same thing and only expects 2 columns.

I locally installed the library and edited the code to the ljspeech() formatter function to pull cols[1] instead and this fixed the error and my model was able to train without issue

[darknight2163]

Follow this 'LJ speech format' for writing on each line of the transcript file: wav_file_name|transcript|transcript
eg. :

LJ001-0002|in being comparatively modern.|in being comparatively modern.
LJ001-0006|And it is worth mention in passing that, as an example of fine typography,|And it is worth mention in passing that, as an example of fine typography,
LJ001-0008|has never been surpassed.|has never been surpassed.
.
.
.
and so on for all the lines of the file.
It worked for me so hope this will work for you too👍

[hrushik98]

This works! Thanks!

[siagupta-03]

(venv) C:\Users\Sia Gupta\EasyEat\TTS>python train.py

Setting up Audio Processor...
| > sample_rate:22050
| > resample:False
| > num_mels:80
| > log_func:np.log10
| > min_level_db:-100
| > frame_shift_ms:None
| > frame_length_ms:None
| > ref_level_db:20
| > fft_size:1024
| > power:1.5
| > preemphasis:0.0
| > griffin_lim_iters:60
| > signal_norm:True
| > symmetric_norm:True
| > mel_fmin:0
| > mel_fmax:None
| > pitch_fmin:1.0
| > pitch_fmax:640.0
| > spec_gain:20.0
| > stft_pad_mode:reflect
| > max_norm:4.0
| > clip_norm:True
| > do_trim_silence:True
| > trim_db:45
| > do_sound_norm:False
| > do_amp_to_db_linear:True
| > do_amp_to_db_mel:True
| > do_rms_norm:False
| > db_level:None
| > stats_path:None
| > base:10
| > hop_length:256
| > win_length:1024
T File "C:\Users\Sia Gupta\EasyEat\TTS\train.py", line 63, in
train_samples, eval_samples = load_tts_samples(
^^^^^^^^^^^^^^^^^
File "C:\Users\Sia Gupta\EasyEat\TTS\TTS\tts\datasets_init_.py", line 120, in load_tts_samples
meta_data_train = formatter(root_path, meta_file_train, ignored_speakers=ignored_speakers)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Sia Gupta\EasyEat\TTS\TTS\tts\datasets\formatters.py", line 201, in ljspeech
text = cols[2]
~~~~^^^
IndexError: list index out of range

I have already cleaned the data with sanity_check and removed the rows but still geting the same cols error kindly help