Original on kaggle

Load frame lengths 

train_frames = {}

total = 0
with open("../input/cmu-us-awb-arctic-fairseq-files/train.tsv") as f:
    for line in f.readlines():
        if not "\t" in line:
            continue
        pieces = line.strip().split("\t")
        total += int(pieces[1])
        assert len(pieces) == 2
        id = pieces[0].replace(".wav", "")
        train_frames[id] = int(pieces[1])

total / 16000

4295.0
4295 / 60

71.58333333333333
MINS = [i * 5 for i in range(1, 13)]

MINS

[5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60]
WAVDIR = "/kaggle/input/ljspeech-for-asr/wav16"

Minute splits 

for min in MINS:
    frames = min * 60 * 16000
    idlist = [k for k in train_frames.keys()]
    outtsv = f"{min}mins.tsv"
    with open(outtsv, "w") as of:
        current = 0
        of.write(f"{WAVDIR}\n")
        while frames > 0 and frames > current:
            id = idlist.pop(0)
            current = train_frames[id]
            of.write(f"{id}.wav\t{current}\n")
            frames = frames - current
        max = 0
        maxid = ""
        for id in idlist:
            time = train_frames[id]
            if time > current:
                continue
            if time > max:
                max = time
                maxid = id
        of.write(f"{maxid}.wav\t{max}\n")

Generate ltr files 

def fairseqify(text):
    text = text.strip().replace("  ", " ")
    words = text.split(" ")
    spread = [" ".join(a) for a in words]
    return " | ".join(spread) + " |"

transcripts = {}
with open("../input/cmu-us-awb-arctic-fairseq-files/text.tsv") as tf:
    for line in tf.readlines():
        line = line.strip()
        if not "\t" in line:
            pass
        parts = line.split("\t")
        assert len(parts) == 2
        transcripts[parts[0]] = fairseqify(parts[1])

import glob
for tsv in glob.glob("*.tsv"):
    out = tsv.replace(".tsv", ".ltr")
    with open(tsv) as inf, open(out, "w") as outf:
        for line in inf.readlines()[1:]:
            id, _ = line.split("\t")
            id = id.replace(".wav", "")
            outf.write(f"{transcripts[id]}\n")

Tidy up 

!wget https://dl.fbaipublicfiles.com/fairseq/wav2vec/dict.ltr.txt

--2022-05-07 20:21:52--  https://dl.fbaipublicfiles.com/fairseq/wav2vec/dict.ltr.txt
Resolving dl.fbaipublicfiles.com (dl.fbaipublicfiles.com)... 172.67.9.4, 104.22.75.142, 104.22.74.142, ...
Connecting to dl.fbaipublicfiles.com (dl.fbaipublicfiles.com)|172.67.9.4|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 207 [text/plain]
Saving to: ‘dict.ltr.txt’

dict.ltr.txt        100%[===================>]     207  --.-KB/s    in 0s      

2022-05-07 20:21:53 (40.9 MB/s) - ‘dict.ltr.txt’ saved [207/207]


!for i in *mins.tsv;do b=$(basename $i ".tsv");mkdir $b; mv $b.tsv $b/train.tsv; mv $b.ltr $b/train.ltr; cp dict.ltr.txt ../input/cmu-us-awb-arctic-fairseq-files/test.* $b/;cp ../input/cmu-us-awb-arctic-fairseq-files/dev.tsv $b/valid.tsv; cp ../input/cmu-us-awb-arctic-fairseq-files/dev.ltr $b/valid.ltr;done