kaldi_python_io_read_write_scp_ark_matrices_vectors.py

1import kaldi_python_io
2import numpy as np
3
4# 1. Read from an scp file (e.g., feats.scp)
5# The ArchiveReader returns a generator of (key, matrix/vector)
6scp_path = "feats.scp"
7with kaldi_python_io.ArchiveReader(scp_path) as reader:
8    for key, mat in reader:
9        print(f"Key: {key}, Matrix Shape: {mat.shape}")
10
11# 2. Write to an ark/scp pair
12# This creates an archive file and a corresponding script file
13ark_scp_path = "ark,scp:out.ark,out.scp"
14with kaldi_python_io.ArchiveWriter(ark_scp_path) as writer:
15    for i in range(10):
16        fake_data = np.random.randn(100, 40).astype(np.float32)
17        writer.write(f"utterance_{i}", fake_data)
18
19# 3. Random access reading (using a script file)
20# This allows fetching specific keys without iterating through the whole file
21script_reader = kaldi_python_io.ScriptReader("out.scp")
22mat = script_reader["utterance_5"]
23print(f"Random Access - Key: utterance_5, Shape: {mat.shape}")