📁 Data Preparation (Speaker Verification - ID)

This guide explains how to prepare training and validation datasets for Speaker Verification (ID) using the soundkit CLI. In this task, speaker-labeled utterances are augmented and organized into TFRecords suitable for training an embedding-based identity recognition model.

---

🔧 Run data Mode

soundkit -t id -m data -c configs/id/id.yaml

---

🧾 Data Parameters

Parameter	Description
path_tfrecord	Output path for storing generated TFRecords. Defaults to ${job_dir}/tfrecords
tfrecord_datalist_name	YAML files listing TFRecord shards used during training/validation (train_tfrecord.yaml, val_tfrecord.yaml)
force_download	If true, forces re-download of the listed datasets
reverb_prob	Probability of applying reverberation during sample generation
num_processes	Number of parallel worker processes for data synthesis
corpora	List of dataset specifications (name, type, split) used for generating speaker-labeled and noise examples. See below for YAML format.
snr_dbs	List of signal-to-noise ratios (in dB) used when mixing speech and noise
target_length_in_secs	Duration of each generated audio clip
min_amp, max_amp	Amplitude range for random scaling of waveform samples
signal	Dictionary of signal-level preprocessing options (e.g., sampling rate, DC removal) applied to audio before feature extraction. See below for YAML format.
num_sentences	Number of utterances sampled per speaker for ID training
ppls_per_group	Number of speakers grouped per batch for training contrastive loss models

---

📦 Corpora Configuration

Corpora specify which datasets are used for generating speaker-labeled examples:

corpora:
  - {name: vad_train-clean-100, type: speech, split: train}
  - {name: vad_train-clean-360, type: speech, split: train}
  - {name: vad_train-other-500, type: speech, split: train}
  - {name: vad_dev-clean, type: speech, split: val}
  - {name: ESC-50-master, type: noise, split: train-val}
  - {name: FSD50K, type: noise, split: train-val}
  - {name: musan, type: noise, split: train-val}
  - {name: wham_noise, type: noise, split: train-val}
  - {name: rirs_noises, type: reverb, split: train-val}

• type: speech: Provides speaker-labeled utterances.
• type: noise: Mixed with speech to create variable SNR conditions.
• type: reverb: Impulse responses for applying environmental reverb.

To register your own dataset, see BYOD.

---

🎚 Signal Preprocessing

Signal-level options defined in the signal section control waveform transformations:

signal:
  sampling_rate: 16000
  dc_removal: true

Parameter	Description
sampling_rate	Target sample rate for all audio
dc_removal	If true, removes DC bias from audio

---

🧪 Output

Running data mode will generate:

• TFRecord Files: Saved in ${job_dir}/tfrecords, e.g., train-00001.tfrecord
• TFRecord Lists: YAML files (train_tfrecord.yaml, val_tfrecord.yaml) that index each split
• These outputs are consumed during the training and evaluation phases of the ID pipeline

---

Ensure you’ve set up your corpora and parameters correctly before launching data preparation. For next steps, see the Train guide.