Overview

The audio framework provides a software mixing engine (audio mixer) for all audio devices, allowing more than one process to play or record audio at the same time.

Multi-Stream Codecs

The audio mixer supports multi-stream Codecs. These devices have DSP engines that provide sample rate conversion, hardware mixing, and other features. The use of such hardware features is opaque to applications.

Backward Compatibility

It is not possible to disable the mixing function. Applications must not assume that they have exclusive access to the audio device.

Audio Formats

Digital audio data represents a quantized approximation of an analog audio signal waveform. In the simplest case, these quantized numbers represent the amplitude of the input waveform at particular sampling intervals. To achieve the best approximation of an input signal, the highest possible sampling frequency and precision should be used. However, increased accuracy comes at a cost of increased data storage requirements. For instance, one minute of monaural audio recorded in u-Law format (pronounced mew-law) at 8 KHz requires nearly 0.5 megabytes of storage, while the standard Compact Disc audio format (stereo 16-bit linear PCM data sampled at 44.1 KHz) requires approximately 10 megabytes per minute.

An audio data format is characterized in the audio driver by four parameters: sample Rate, encoding, precision, and channels. Refer to the device-specific manual pages for a list of the audio formats that each device supports. In addition to the formats that the audio device supports directly, other formats provide higher data compression. Applications can convert audio data to and from these formats when playing or recording.

Sample Rate

Sample rate is a number that represents the sampling frequency (in samples per second) of the audio data.

The audio mixer always configures the hardware for the highest possible sample rate for both play and record. This ensures that none of the audio streams require compute-intensive low pass filtering. The result is that high sample rate audio streams are not degraded by filtering.

Sample rate conversion can be a compute-intensive operation, depending on the number of channels and a device's sample rate. For example, an 8KHz signal can be easily converted to 48KHz, requiring a low cost up sampling by 6. However, converting from 44.1KHz to 48KHz is computer intensive because it must be up sampled by 160 and then down sampled by 147. This is only done using integer multipliers.

Applications can greatly reduce the impact of sample rate conversion by carefully picking the sample rate. Applications should always use the highest sample rate the device supports. An application can also do its own sample rate conversion (to take advantage of floating point and accelerated instructions) or use small integers for up and down sampling.

All modern audio devices run at 48 kHz or a multiple thereof, hence just using 48 kHz can be a reasonable compromise if the application is not prepared to select higher sample rates.

Encodings

An encoding parameter specifies the audiodata representation. u-Law encoding corresponds to CCITT G.711, and is the standard for voice data used by telephone companies in the United States, Canada, and Japan. A-Law encoding is also part of CCITT G.711 and is the standard encoding for telephony elsewhere in the world. A-Law and u-Law audio data are sampled at a rate of 8000 samples per second with 12-bit precision, with the data compressed to 8-bit samples. The resulting audio data quality is equivalent to that of standard analog telephone service.

Linear Pulse Code Modulation (PCM) is an uncompressed, signed audio format in which sample values are directly proportional to audio signal voltages. Each sample is a 2's complement number that represents a positive or negative amplitude.

Precision

Precision indicates the number of bits used to store each audio sample. For instance, u-Law and A-Law data are stored with 8-bit precision. PCM data can be stored at various precisions, though 16-bit is the most common.

Channels

Multiple channels of audio can be interleaved at sample boundaries. A sample frame consists of a single sample from each active channel. For example, a sample frame of stereo 16-bit PCM data consists of 2 16-bit samples, corresponding to the left and right channel data. The audio mixer sets the hardware to the maximum number of channels supported. If a mono signal is played or recorded, it is mixed on the first two (usually the left and right) channel only. Silence is mixed on the remaining channels.

Supported Formats

The audio mixer supports the following audio formats:

Encoding            Precision  Channels
Signed Linear PCM   32-bit     Mono or Stereo
Signed Linear PCM   16-bit     Mono or Stereo
Signed Linear PCM   8-bit      Mono or Stereo
u-Law               8-bit      Mono or Stereo
A-Law               8-bit      Mono or Stereo

The audio mixer converts all audio streams to 24-bit Linear PCM before mixing. After mixing, conversion is made to the best possible Codec format. The conversion process is not compute intensive and audio applications can choose the encoding format that best meets their needs.

The mixer discards the low order 8 bits of 32-bit Signed Linear PCM in order to perform mixing. (This is done to allow for possible overflows to fit into 32-bits when mixing multiple streams together.) Hence, the maximum effective precision is 24-bits.