To boost wider use of its 16-bit digital signal controller in embedded audio applications, Microchip is making available a royalty-free speech encoding (compression) and decoding (decompression) library for its dsPIC family.
The software, and the dsPIC devices, is targeted at applications such as voice over IP, tapeless answering machines, voice announcement systems, medical equipment, appliances, building and home safety systems, intercoms, and proprietary communications equipment.
According to Sumit Mitra, vice president of Microchip's Digital Signal Controller division, the dsPIC30F Speech Encoding/Decoding Library is
based on the Speex open source compression format, and achieves a 16:1
compression ratio and toll quality speech (PESQ 3.7-4.2).
He said it is designed to allow embedded designers to build systems that can transmit more messages over a given bandwidth, or store longer amounts of speech in a given memory space, while maintaining excellent sound quality and intelligibility.
"Many design engineers are seeking to add speech compression to their
bandwidth- or memory-constrained applications, but have opted for lower
compression methodologies to avoid a stiff per-unit royalty for the
intellectual property," said Mitra. "By adapting the open-source Speex audio compression format, we are providing users with low-cost, high-quality compression capability from the open source free software community.
He said any embedded application that transmits voice over limited bandwidth or
has memory constrained playback can benefit from this speech compression/ decompression library.
The library has a well-documented Application Programmer's Interface (API) that makes it easy to incorporate speech compression and/or decompression
into an application. The library also supports optional voice activity
detection.
The speech encoder samples data at 8 kHz, using either the dsPIC DSC's
on-chip 12-bit analog-to-digital converter or a 16-bit input obtained from an external CODEC, via the on-chip codec interface port. The encoder
achieves up to a 16:1 compression ratio and generates a fixed 8 kbps
output rate.
Output of decoded speech can be accomplished using the dsPIC DSC's
standard PWM or via the interface to an external CODEC. The encoder
requires 19 MIPS (worst case), 33 Kbytes of program memory and 6.2 Kbytes
of RAM. The decoder requires 3 MIPS, 15 Kbytes of program memory, and 3.2 Kbytes of RAM.
Storing speech requires 1 Kbyte of memory for each second of speech. The
library permits the compressed speech data to be stored in on-chip memory
(Flash, EEPROM or RAM) or to an external Flash memory.
Playback-only applications can benefit from a PC-based speech encoder
utility that is included on the library CD. This utility creates encoded
speech files from a PC microphone or WAV files.
To evaluate the library, developers can make use of the dsPICDEM 1.1 General Purpose Development Board (DM300014). All licensing options for the dsPIC30F Speech Encoding/Decoding Library are immediately available. Among these options is a one-year evaluation/development license, which is unaltered from the library that is used for production and can be purchased online at the company's web site.
The evaluation/development version of the dsPIC30F Speech Encoding/Decoding Library (SW300070-EVAL) is $5USD. Industry-unique, one-time-fee licensing options are also available for production, which Mitra said, is counter to the per-unit royalty business model that is common for this class of library. The license fee for systems based on the algorithms start at $2,500 for 5,000-unit production.
Speex is copyrighted by Jean-Marc Valin and the Xiph.Org Foundation and is also available under the Xiph.org variant of the BSD license. For more information about Speex, go to www.speex.org/fsos.html
Microchip Technology, Inc.
Chandler, Az.
www.microchip.com
