Electronicdesign 2437 Xl speech

Microcontroller Talks—And Listens

March 25, 2010
Sensory's NLP-5x is a 16-bit DSP designed for speech chores. It handles speaker-dependent and -independent recognition

80-MHz NLP-5x microcontroller

The NLP-5x microcontroller from Sensory not only generates speech from text, it also handles speaker-independent and speaker-dependent voice input. The voice input is limited to short phrases to improve accuracy, so leave dictation to a PC. Still, the NLP-5x chip is ideal for a range of applications including voice-activated appliances.

The 80-MHz NLP-5x (see the figure) is based on a 16-bit DSP core tailored for voice processing. Combined with Sensory’s FluentChip firmware, the chip can handle up to 750 seconds of compressed text-to-speech (TTS) without using off-chip memory. It can also store multiple speaker-dependent vocabularies as well as speaker-independent vocabularies.


The voice recognition system is about 95% accurate for voice-independent interaction. It can recognize multiple phrases. Also, it can handle user training with voice interaction for speaker-dependent recognition. The choice of words is key to improving accuracy, and Sensory works with most customers to optimize their vocabulary. Changing one or two keywords often can significantly improve recognition performance.

Developers have a choice for audio output. A simple pulse-width modulator (PWM) can drive a low-end speaker directly. The two 16-bit digital-to-analog converters (DACs) are designed for stereo output including speech and sound effects using a 48-kHz sample rate.

The software also supports 24-voice, MIDI-compatible, stereo music synthesis.  The MIDI support operates in parallel to voice support, allowing the two outputs to be mixed.

Audio output can be synchronized with external devices like motors. As a result, the mouth movements of toys can be linked to the speech output. Likewise, the system supports beat detection from the audio input. This primarily targets toys where the user beats out a rhythm. All this support as well as the voice recognition support is integrated into Sensory’s scripting language, which greatly simplifies application development.


The NLP-5x is designed to be more than a voice processing system. It has plenty of headroom and peripherals to handle many application-oriented chores including motor control for up to three motors. The infrared (IR) support allows it to work with a remote control.

Most of the digital peripherals and 3.6-V tolerant I/O ports are the same as those found on other microcontrollers and digital signal controllers (DSCs) in this realm.

Overall, Sensory’s NLP-5x line represents a powerful, low-cost, single-chip solution for voice-related man-machine interfaces. Pricing starts at $2. A development kit costs $1500.


Sponsored Recommendations


To join the conversation, and become an exclusive member of Electronic Design, create an account today!