原始PCM音频存储格式及FFmpeg转换后采样表示问询

Hey there! Let's tackle your two questions about raw PCM and FFmpeg's output—they're super relevant for audio processing apps, so great questions to clarify upfront.

Raw PCM is the most basic form of audio storage—it's pure, uncompressed sample data with no extra frills attached. Here's a breakdown of how it works:

• No headers, metadata, or frame structures: The file contains nothing but sequential audio samples. There's no info about sample rate, channel count, or bit depth stored in the file itself—you have to track that context separately (since raw PCM has no way to self-describe).
• Sample encoding: Each sample represents the audio wave's amplitude at a specific timestamp. The format (like 16-bit signed, 32-bit float) dictates how each amplitude is stored as bytes. For example, 16-bit signed PCM uses 2 bytes per sample to represent values from -32768 to 32767.
• Multi-channel interleaving: For stereo or surround sound, samples are interleaved. Stereo follows the order: Left sample 1 → Right sample 1 → Left sample 2 → Right sample 2 → ...—no gaps or separate blocks for each channel.

Your command ffmpeg -i test.wav -f s16le -acodec pcm_s16le output.dat is generating a raw PCM file, so let's break down exactly what's in it:

• No frame structure or headers: FFmpeg strips away all the WAV container data (like the RIFF header that includes sample rate, channel count, etc.) and writes only the raw sample bytes. Every byte in output.dat is part of an audio sample—there's no extra framing, sync markers, or metadata here.
• Sample representation: As you noted, pcm_s16le is 16-bit signed little-endian PCM. Each sample takes 2 bytes, and stereo channels are interleaved in the L1 R1 L2 R2... order. To convert these to floating-point values for your app, you'd take each 2-byte signed integer (range: -32768 to 32767) and divide by 32768.0 to get a float in the [-1.0, 1.0] range.
• If you ever need to verify, you can use FFmpeg to probe the file (though raw PCM doesn't have self-description, you can pass the known parameters with ffmpeg -f s16le -ar 44100 -ac 2 -i output.dat -vn -f null - to check if it plays back correctly, replacing the sample rate and channel count with your input WAV's values).

内容的提问来源于stack exchange，提问作者Tian Ye