Python读取WAV文件：采样率与数据维度不匹配及波形频谱图生成需求

Hey there! Let's work through that dimension mismatch issue you're facing when generating the spectrogram. Here's a breakdown of what's likely going on and how to fix it:

First, Let's Diagnose the Problem #

From your output, I can see your samples variable is a 1-dimensional array (shape: (14466512,)), which makes sense for a mono WAV file. Most spectrogram functions (custom or library-provided) expect either:

• A 2-dimensional array (shape: (number_of_samples, number_of_channels)) for multi-channel audio, or
• A 1-dimensional array specifically for mono audio, but sometimes functions are built to handle multi-channel inputs by default.

The "dimension mismatch" error is almost certainly because your log_specgram function is expecting a 2D array, but you're passing a 1D array.

Step-by-Step Fixes #

1. Check and Adjust the Sample Array Dimension

First, confirm the shape of your samples to be sure:

print(samples.shape)  # Should output (14466512,) for mono audio

If it's 1D, convert it to a 2D array (since mono audio is just 1 channel):

import numpy as np

# Reshape to (number_of_samples, 1) for single-channel audio
samples = samples.reshape(-1, 1)

2. Convert to Float and Normalize (Critical for Most Spectrogram Functions)

wavfile.read() returns integer values (usually int16), but many spectral analysis functions expect floating-point data normalized to the range [-1, 1]. Fix this with:

# Convert to float32 and normalize
samples = samples.astype(np.float32) / np.iinfo(np.int16).max

3. Verify Function Parameter Order

Double-check that your log_specgram function expects (samples, sample_rate) as inputs. Some functions reverse the order (e.g., (sample_rate, samples)). If you're using a custom log_specgram, confirm its definition matches how you're calling it.

Full Working Example (Waveform + Spectrogram) #

Here's a complete code snippet that includes both plots, assuming you're using standard libraries like numpy, scipy, and matplotlib:

import numpy as np
from scipy.io import wavfile
from scipy import signal
import matplotlib.pyplot as plt

# Custom log_specgram implementation (adjust if yours is different)
def log_specgram(audio, sample_rate, window_size=20, step_size=10, eps=1e-10):
    nperseg = int(round(window_size * sample_rate / 1000))
    noverlap = int(round(step_size * sample_rate / 1000))
    freqs, times, spec = signal.spectrogram(
        audio, fs=sample_rate, nperseg=nperseg, noverlap=noverlap
    )
    return freqs, times, np.log(spec.T.astype(np.float32) + eps)

# Load your audio file
sample_rate, samples = wavfile.read(files[:1][0])
print(f"Sample Rate: {sample_rate}")
print(f"Original Samples Shape: {samples.shape}")

# Fix dimension and data type
if len(samples.shape) == 1:
    samples = samples.reshape(-1, 1)
samples = samples.astype(np.float32) / np.iinfo(np.int16).max

# Generate spectrogram using the first (and only) channel
freqs, times, spectrogram = log_specgram(samples[:, 0], sample_rate)

# Plot Waveform
plt.figure(figsize=(12, 8))
plt.subplot(2, 1, 1)
plt.plot(np.arange(len(samples)) / sample_rate, samples[:, 0])
plt.title("Audio Waveform")
plt.xlabel("Time (seconds)")
plt.ylabel("Amplitude")

# Plot Spectrogram
plt.subplot(2, 1, 2)
plt.imshow(
    spectrogram.T,
    aspect="auto",
    origin="lower",
    extent=[times.min(), times.max(), freqs.min(), freqs.max()]
)
plt.title("Log Spectrogram")
plt.xlabel("Time (seconds)")
plt.ylabel("Frequency (Hz)")
plt.colorbar(label="Log Amplitude")

plt.tight_layout()
plt.show()

Additional Notes #

• If your log_specgram is from a library like librosa, it likely expects a 1D array for mono audio. In that case, skip the reshape step and just pass samples directly (after normalizing to float).
• If you still get errors, print the full traceback—it will tell you exactly which dimension the function is expecting vs. what you're passing.

内容的提问来源于stack exchange，提问作者bytebiscuit