[recipe] refine yaml for librispeech

[xingchensong]

follow #2205

[xingchensong]

librispeech 模型转换脚本（sos id从vocabsize - 1 变为 2）

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
# Copyright [2023-12-13] <[email protected], Xingchen Song>

import torch

old_state = torch.load('/mnt/d/BaiduSyncdisk/downloads/ckpt/20210610_u2pp_conformer_exp_librispeech/final.pt')
new_state = {}
change_list = ['decoder.left_decoder.output_layer.weight',
               'decoder.left_decoder.output_layer.bias',                                                                                                                                                                                            'decoder.left_decoder.embed.0.weight',
               'decoder.right_decoder.output_layer.weight',
               'decoder.right_decoder.output_layer.bias',
               'decoder.right_decoder.embed.0.weight',
               'ctc.ctc_lo.weight',
               'ctc.ctc_lo.bias']
for key in old_state.keys():
    if key in change_list:
        print("processing {}, {}".format(key, old_state[key].size()))
        tensor = old_state[key]
        new_tensor = torch.zeros_like(tensor)
        if len(tensor.size()) == 2:  # weight
            new_tensor[:2, :] = tensor[:2, :]
            new_tensor[2, :] = tensor[-1, :]
            new_tensor[3:, :] = tensor[2:-1, :]
        elif len(tensor.size()) == 1:  # bias
            new_tensor[:2] = tensor[:2]
            new_tensor[2] = tensor[-1]
            new_tensor[3:] = tensor[2:-1]
        else:
            raise NotImplementedError
        new_state[key] = new_tensor
    elif "concat_linear" in key:
        continue
    else:
        new_state[key] = old_state[key]

torch.save(new_state, "/mnt/d/BaiduSyncdisk/downloads/ckpt/20210610_u2pp_conformer_exp_librispeech/final.sos2.pt")

old_units = '/mnt/d/BaiduSyncdisk/downloads/ckpt/20210610_u2pp_conformer_exp_librispeech/units.txt'
new_units = '/mnt/d/BaiduSyncdisk/downloads/ckpt/20210610_u2pp_conformer_exp_librispeech/units.sos2.txt'

with open(old_units, "r") as fin, open(new_units, "w") as fout:
    lines = fin.readlines()
    fout.write(lines[0])
    fout.write(lines[1])
    fout.write("<sos/eos> 2\n")
    for line in lines[2:-1]:
        line = line.strip().split()
        token, token_id = line[0], line[1]
        fout.write("{} {}\n".format(token, int(token_id) + 1))

解码通过

[xingchensong]

修改后结果对的上