Model reference · open weights

xun_tal_zh_adult_teach_branchformer

Audio espnet Speech→text 1 build Open weights 9 dl/mo

xun_tal_zh_adult_teach_branchformer is an open-weight audio or speech model from espnet. xun_tal_zh_adult_teach_branchformer (BF16) weighs 391 MB; the smallest configuration that runs it is RTX 3060 12 GB.

  • xun_tal_zh_adult_teach_branchformer is an automatic speech recognition model developed by espnet for the Chinese language.
  • It is trained on the tal_zh_adult_teach recipe and operates at a 16 kHz sampling rate.
  • The model is distributed under the cc-by-4.0 license.

Summary of the espnet/xun_tal_zh_adult_teach_branchformer model card, 2026-10-01

What it is

Released byespnet
Released2025-12-17
VRAM391 MB for the weights

What it runs on

Memory and cards for xun_tal_zh_adult_teach_branchformer (BF16)

391 MBweights, file size
1.6 GBruntime overhead
CardOne streamMemory
RTX 3060 12 GBfits11.6 GB
RTX 4060 Ti 16 GBfits15.4 GB
RTX 3090 24 GBfits23.4 GB
RTX 4090 24 GBfits23.4 GB
RTX 5090 32 GBfits31.0 GB
L40S 48 GBfits44.0 GB
A100 80 GBfits78.2 GB
H100 80 GBfits78.1 GB
RTX PRO 6000 Blackwell 96 GBfits93.8 GB
DGX Spark (GB10) 128 GB unifiedfits107 GB
H200 141 GBfits138 GB
B200 180 GBfits176 GB

From the model card

What espnet says about xun_tal_zh_adult_teach_branchformer

Read the model card

ESPnet2 ASR model

espnet/xun_tal_zh_adult_teach_branchformer

This model was trained by Xun Gong using tal_zh_adult_teach recipe in espnet.

Demo: How to use in ESPnet2

Follow the ESPnet installation instructions if you haven't done that already.

cd espnet
git checkout c61e21170895255cb93d1d3857bc15b818daca99
pip install -e .
cd egs2/tal_zh_adult_teach/asr1
./run.sh --skip_data_prep false --skip_train true --download_model espnet/xun_tal_zh_adult_teach_branchformer

RESULTS

Environments

  • date: Wed Dec 17 15:20:29 EST 2025
  • python version: 3.10.19 (main, Oct 21 2025, 16:43:05) [GCC 11.2.0]
  • espnet2 version: espnet2 202511
  • pytorch version: pytorch 2.6.0+cu126
  • Git hash: c61e21170895255cb93d1d3857bc15b818daca99
    • Commit date: Mon Dec 15 04:37:35 2025 -0800

exp/asr_train_asr_branchformer_e24_amp_raw_zh_char_sp

WER

datasetSntWrdCorrSubDelInsErrS.Err
decode_asr_branchformer_asr_model_valid.acc.ave/test6072850038.859.12.11.362.578.6
decode_asr_branchformer_lm_lm_train_lm_transformer_zh_char_valid.loss.ave_asr_model_valid.acc.ave/test6072850038.959.21.81.762.878.5
decode_asr_branchformer_asr_model_valid.acc.ave/org/dev3208429141.256.32.51.660.572.8
decode_asr_branchformer_lm_lm_train_lm_transformer_zh_char_valid.loss.ave_asr_model_valid.acc.ave/org/dev3208429141.556.12.42.160.672.5

CER

datasetSntWrdCorrSubDelInsErrS.Err
decode_asr_branchformer_asr_model_valid.acc.ave/test607216005789.48.22.40.911.478.6
decode_asr_branchformer_lm_lm_train_lm_transformer_zh_char_valid.loss.ave_asr_model_valid.acc.ave/test607216005789.28.22.71.111.978.5
decode_asr_branchformer_asr_model_valid.acc.ave/org/dev32087733491.37.21.50.89.672.8
decode_asr_branchformer_lm_lm_train_lm_transformer_zh_char_valid.loss.ave_asr_model_valid.acc.ave/org/dev32087733490.97.41.71.010.172.5

ASR config

config: conf/train_asr_branchformer_e24_amp.yaml
print_config: false
log_level: INFO
drop_last_iter: false
dry_run: false
iterator_type: sequence
valid_iterator_type: null
output_dir: exp/asr_train_asr_branchformer_e24_amp_raw_zh_char_sp
ngpu: 1
seed: 0
num_workers: 4
num_att_plot: 3
dist_backend: nccl
dist_init_method: env://
dist_world_size: 4
dist_rank: 0
local_rank: 0
dist_master_addr: localhost
dist_master_port: 51233
dist_launcher: null
multiprocessing_distributed: true
unused_parameters: false
sharded_ddp: false
use_deepspeed: false
deepspeed_config: null
gradient_as_bucket_view: true
ddp_comm_hook: null
cudnn_enabled: true
cudnn_benchmark: false
cudnn_deterministic: true
use_tf32: false
collect_stats: false
write_collected_feats: false
max_epoch: 60
patience: null
val_scheduler_criterion:
- valid
- acc
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
-   - valid
    - acc
    - max
keep_nbest_models: 10
nbest_averaging_interval: 0
grad_clip: 5
grad_clip_type: 2.0
grad_noise: false
accum_grad: 1
no_forward_run: false
resume: true
train_dtype: float32
use_amp: true
log_interval: null
use_matplotlib: true
use_tensorboard: true
create_graph_in_tensorboard: false
use_wandb: false
wandb_project: null
wandb_id: null
wandb_entity: null
wandb_name: null
wandb_model_log_interval: -1
detect_anomaly: false
use_adapter: false
adapter: lora
save_strategy: all
adapter_conf: {}
pretrain_path: null
init_param: []
ignore_init_mismatch: false
freeze_param: []
num_iters_per_epoch: null
batch_size: 20
valid_batch_size: null
batch_bins: 25000000
valid_batch_bins: null
category_sample_size: 10
upsampling_factor: 0.5
category_upsampling_factor: 0.5
dataset_upsampling_factor: 0.5
dataset_scaling_factor: 1.2
max_batch_size: null
min_batch_size: 1
train_shape_file:
- exp/asr_stats_raw_zh_char_sp/train/speech_shape
- exp/asr_stats_raw_zh_char_sp/train/text_shape.char
valid_shape_file:
- exp/asr_stats_raw_zh_char_sp/valid/speech_shape
- exp/asr_stats_raw_zh_char_sp/valid/text_shape.char
batch_type: numel
valid_batch_type: null
fold_length:
- 51200
- 150
sort_in_batch: descending
shuffle_within_batch: false
sort_batch: descending
multiple_iterator: false
chunk_length: 500
chunk_shift_ratio: 0.5
num_cache_chunks: 1024
chunk_excluded_key_prefixes: []
chunk_default_fs: null
chunk_max_abs_length: null
chunk_discard_short_samples: true
train_data_path_and_name_and_type:
-   - dump/raw/train_sp/wav.scp
    - speech
    - kaldi_ark
-   - dump/raw/train_sp/text
    - text
    - text
valid_data_path_and_name_and_type:
-   - dump/raw/dev/wav.scp
    - speech
    - kaldi_ark
-   - dump/raw/dev/text
    - text
    - text
multi_task_dataset: false
allow_variable_data_keys: false
max_cache_size: 0.0
max_cache_fd: 32
allow_multi_rates: false
valid_max_cache_size: null
exclude_weight_decay: false
exclude_weight_decay_conf: {}
optim: adam
optim_conf:
    lr: 0.001
    weight_decay: 1.0e-06
scheduler: warmuplr
scheduler_conf:
    warmup_steps: 35000
token_list:
-
-
- 的
- 一
- 是
- 这
- 个
- 啊
-
- 我
- 们
- 就
- 了
- 那
- 不
- 对
- 来
- 以
- 呢
- 有
- 十
- 你
- 么
- 二
- 吧
- 三
- 好
- 分
- 看
- 它
- 下
- 题
- 要
- 说
- 在
- 之
- 数
- 于
- 可
- 大
- 四
- 后
- 到
- 加
- 能
- 所
- 五
- 上
- 等
- 哎
- 小
- 去
- 还
- 面
- 什
- 时
- 第
- 样
- 也
- 六
- 多
- 点
- 两
- 写
- 角
- 里
- 道
- 他
- 家
- 把
- 方
- 出
- 然
- 都
- 应
- 没
- 减
- 再
- 法
- 给
- 人
- 问
- 为
- 最
- 该
- 儿
- 和
- 比
- 如
- 做
- 用
- 学
- 想
- 会
- 百
- 八
- 果
- 同
- 乘
- 边
- 算
- 定
- 成
- 得
- 先
- 间
- 现
- 接
- A
- 零
- 话
- 少
- 呀
- X
- 七
- 咱
- 过
- 子
- 九
- 实
- 怎
- 候
- 知
- 其
- 吗
- 中
- 种
- 直
- 派
- 变
- 求
- 老
- 式
- 意
- 正
- 很
- 位
- 度
- 只
- 几
- 音
- 号
- 赛
- B
- 自
- 师
- 形
- 行
- 因
- 除
- 常
- 关
- 平
- 相
- 解
- 当
- 些
- 完
- 友
- 朋
- 起
- 值
- 讲
- 倍
- 但
- 答
- 前
- 天
- 开
- 尔
- 长
- 特
- 非
- 嗯
- 根
- 着
- 阿
- 作
- 发
- 结
- 单
- 理
- 哪
- C
- 选
- 表
- 文
- 线
- 别
- 己
- 部
- 次
- 谁
- 画
- 刚
- 跟
- 考
- 记
- 系
- 回
- 字
- 叫
- 情
- 整
- 思
- 负
- 经
- 找
- 明
- 注
- 课
- 从
- 错
- 图
- 地
- K
- 又
- 扣
- 告
- 已
- 条

Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.

How it works

How audio & music work

Audio or textinputAudio modelrecognise / synthesiseText or audiooutputSpeech-to-text turns audio into text; text-to-speech and music models turn text into audio.
© 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms