TWELP 480 bps Vocoder

Details

WARNING!
This page contains information about the previous version (4.0).
The current version (5.0) provides significantly improved speech quality.
We will update the information in the near future.

(version 5.0) provides uniquely high speech quality and intelligibility for such a very low bit rate — surpassing the MELPe at 600 bps.

For Digital HF Radio and other markets.

TWELP Technology Features. The vocoder is based on the newest technology of speech coding called "Tri-Wave Excited Linear Prediction" (TWELP) that was developed by experts of DSPINI. 

TWELP technology is a new class of vocoders that differs from any other LPC-based vocoders in:

  • an advance reliable method of pitch estimation
  • a pitch-synchronous analysis
  • an advance tri-wave model of excitation
  • newest quantization schemes
  • a pitch-synchronous synthesis

Thanks to these unique features, TWELP technology provides a much better speech quality in comparison with any well-known technologies, including AMBE+2, MELPe, ACELP, etc. at the same bit rate in the range from 300 bps up to 9600 bps and more. Moreover, in contrast to other LBR vocoders (like MELPe, etc.) TWELP provides a much better quality for non-speech signals like sirens, background music, etc.

Superiority In Speech Quality. This is a comparison with the MELPe vocoder, which operates at 600 bps.
The TWELP 480 bps and MELPe 600 bps vocoders were tested using the ITU-T P.50 speech base in 20 different languages.
The speech base was updated by removing all non-speech pauses.
The ITU-T P.862 tool was used to evaluate speech quality in terms of PESQ scores:

 
The diagram demonstrates superiority of TWELP 480 over MELPe 600 in a speech quality. Exact numbers are shown in the table below.
LanguageTWELP 480 bpsMELPe 600 bps
American 2.393 2.211
Arabic 2.323 2.168
British 2.409 2.270
Chinese 2.330 2.055
Danish 2.405 2.174
Dutch 2.316 2.145
Finnich 2.340 2.243
French 2.458 2.258
German 2.478 2.227
Greek 2.411 2.164
Hindi 2.490 2.358
Hungarian 2.434 2.293
Italian 2.628 2.417
Japanese 2.421 2.308
Norwegian 2.373 2.197
Polish 2.436 2.274
Portuguese 2.512 2.370
Russian 2.331 2.119
Spanish 2.498 2.322
Swedish 2.578 2.437
Average2.4282.251

Superiority of the TWELP 480 is on average 0.177 PESQ

 Speech Intelligibility. Here is the comparison with MELPe vocoder, operating at much higher bit rate 600 bps. TWELP 480 bps vocoder and MELPe 600 bps vocoder were tested, using ITU-T P.50 speech base for 20 different languages.
STOI (Short-Time Objective Intelligibility) and ESTOI (Extended Short-Time Objective Intelligibility) metrics were used to estimate speech intelligibility: 

 
A diagram just demonstrates a difference in speech intelligibility for TWELP 480 and MELPe 600 vocoders in the STOI metric. Exact numbers are shown in the table below.
LanguageTWELP 480 bpsMELPe 600 bps
American 81.16 79.24
Arabic 78.47 78.50
British 78.41 75.67
Chinese 79.41 77.40
Danish 81.16 79.12
Dutch 79.59 77.04
Finnich 75.54 74.76
French 80.42 78.78
German 80.66 79.00
Greek 79.41 77.89
Hindi 79.66 78.35
Hungarian 80.13 78.14
Italian 79.14 78.03
Japanese 79.98 79.30
Norwegian 80.76 79.25
Polish 80.08 78.37
Portuguese 80.19 78.04
Russian 78.43 75.74
Spanish 78.49 77.82
Swedish 77.66 76.89
Average79.4077.90

A difference is on average 1.50 %
Considering that a low-bitrate vocoder is a nonlinear device that significantly distorts the spectrum of the original speech signal, the ESTOI metric provides more accurate assessments of speech intelligibility after vocoding:
 
A diagram just demonstrates a difference in speech intelligibility for TWELP 480 and MELPe 600 vocoders in the ESTOI metric. Exact numbers are shown in the table below.
LanguageTWELP 480 bpsMELPe 600 bps
American 71.36 69.76
Arabic 70.26 70.05
British 69.39 67.53
Chinese 72.54 69.85
Danish 72.24 69.81
Dutch 70.59 69.58
Finnich 66.82 66.23
French 71.73 70.28
German 70.88 68.55
Greek 72.05 70.59
Hindi 68.70 67.87
Hungarian 69.44 67.46
Italian 69.46 68.60
Japanese 72.31 71.22
Norwegian 73.18 70.59
Polish 71.83 70.79
Portuguese 71.21 69.53
Russian 69.37 66.88
Spanish 70.79 70.33
Swedish 67.06 67.18
Average70.6069.10

A difference is on average 1.50 %
You can download the P.862 and STOI/ESTOI utilities, along with all speech samples, by using the links in the 'Downloads' section at the bottom of the page, and then check all the numbers presented above.


Speech Samples (WAV-files). A few independent experts listened to the TWELP 480 bps vocoder in comparison with the MELPe 600 bps vocoder, using a preference-based method.
The majority of experts preferred TWELP over MELPe, noting a much more natural, human-sounding voice quality in the TWELP vocoder.

You can listen to short samples of the original speech as well as speech processed by both vocoders for any of the 20 languages using the links in the table below.

Additionally, you can download the complete set of P.50 samples as zip files for all languages at once by using the links in the "Downloads" section at the bottom of the page.

LanguageSource speechMELPe 600 bpsTWELP 480 bps
American
Arabic
British
Chinese
Danish
Dutch
Finnich
French
German
Greek
Hindi
Hungarian
Italian
Japanese
Norwegian
Polish
Portuguese
Russian
Spanish
Swedish

Superiority In Quality Of The Non-speech Signals. In contrast to other LBR vocoders (MELPe, AMBE+2, etc.), TWELP vocoders provide a high quality of non-speech signals, including police, ambulance, fire sirens, etc. This feature in conjunction with a high quality natural human-sounding of voice makes TWELP vocoders well suitable for replacement of an analog radio by a digital radio and also for other applications where a high quality transmission of non-speech signals is relevant along with a high quality transmission of speech signals.

Source typeSource signalMELPe 600 bpsTWELP 480 bps
Siren only
With voice

High Robustness To Acoustic Noise. In contrast to other LBR vocoders, TWELP vocoders are well robust to acoustic noise thanks to a robust reliable method of a pitch estimation and other features of TWELP technology.

Moreover, vocoder includes in-built Noise Cancellation—Speech Enhancement (NCSE) functionality that improves a speech quality in a noisy acoustic environment.

NCSE ModeSource signalMELPe 600 bpsTWELP 480 bps
Disabled
Enabled

High Robustness To The Channel Errors. The diagram and the table below show a dependence of the averaged speech quality for AWGN-noisy channel on a different BER in comparison with other vocoders.

Special "robust" versions of the TWELP vocoders include FEC that is integrated with the vocoder on base of "joint source-channel coding" approach that provides a high speech quality simultaneously in a noisy channel as well as in a noiseless channel. FEC can operate with "soft decisions" as well as with "hard decisions" from a modem. "Soft decisions" mode provides a much better robustness in comparison with the "hard decisions" mode. 

 
BER %MELPe 600TWELP 600 RobustTWELP 480
0.00 2.313 2.383 2.389
0.10 2.251 2.367 2.279
0.20 2.205 2.344 2.209
0.30 2.167 2.33 2.153
0.40 2.104 2.293 2.094
0.50 2.069 2.276 2.046
0.60 2.007 2.257 1.994
0.70 1.971 2.243 1.920
0.80 1.925 2.214 1.893
0.90 1.880 2.194 1.844
1.00 1.845 2.152 1.816

Additional Functionalities. The following additional functionalities are developed by DSPINI and integrated into TWELP vocoders:

  • Automatic Gain Control (AGC),
  • Noise Cancellation for Speech Enhancement (NCSE)
  • Voice Activity Detector (VAD),
  • Tone Detection/Generation (Single tones and Dual tones). The tones are transmitted by the vocoder facilities.

Each functionality has unique features, performance and characteristics, providing a significant superiority over any well-known implementations on the market.

Technical Characteristics And Resource Requirements:

Technical characteristics
Bit Rate
(bps)		AlgorithmFrame size
(ms)		Algorithmic delay
(including frame size)
(ms)		Sampling rate
(kHz)		Signal formatBit stream format
480 TWELP 100 120 8 Linear
16-bit
PCM		 48
Additional functionalities
NameFunctionalityTechnical characteristics
NameValue
AGC Automatic Gain 
Control		 Control range: 0 ... +20 dB
NCSE Noise Canceller -
Speech Enhancer		 SNR increasing > 6 dB
Speech quality
improvement		 > 0.1 PESQ
Tone
Detector		 Single/Dual tones 
detection		 In accordance with the international standards
Tone
Generator		 Single/Dual tones 
generation		 Special generator, kept continuity of a signal 
(phase and amplitude of signal of previous frame)
VAD Voice Activity 
Detection		 Reliable detection speech 
in a background noise
CNG Comfort Noise 
Generation		 Type of noise "white"
Level - 60 dB

Resources for ARM Cortex-M4 platform
ModuleMIPS*
peak		Memory (KBytes)
ProgramData
ConstantsChannelHeapStack
Voice Encoder 56.3 38 153 4.6 5.0 1.0
NCSE 6.5
AGC 0.2
Voice Decoder 13.7
Voice Encoder +
Voice Decoder		 70.0
Total 76.7


Resources for TI's C64 DSP platform
ModuleMIPS*
peak		Memory (KBytes)
ProgramData
ConstantsChannelHeapStack
Voice Encoder 20.1 72 153 4.6 5.0 1.0
NCSE 2.8
AGC 0.1
Voice Decoder 3.9
Voice Encoder +
Voice Decoder		 24.0
Total 26.9

Resources (estimated) for TI's C55 DSP platform
ModuleMIPS*
peak		Memory (KBytes)
ProgramData
ConstantsChannelHeapStack
Voice Encoder 34 23 153 4.6 5.0 1.0
NCSE 6.7
AGC 0.2
Voice Decoder 10.0
Voice Encoder +
Voice Decoder		 44
Total 50.9

* DSPINI continues optimization of the TWELP algorithm and the code in order to minimize a computational complexity of the vocoder.

Vulnerability / Security. DSPINI guarantees an ABSOLUTE cleanliness of the software from any undocumented features, undeclared capabilities, etc. All our customers can be sure that none of our software/ code contains secret functions or hiiden features. We are ready to provide the source codes of our software products for an appropriate certification, if needed.

Guarantee And Support. DSPINI guarantees a quality and accordance of all technical characteristics of the product to requirement of the current specifications. Testing and other method of quality control are used for a guarantee support.

Any Platforms. DSPINI can port this vocoder software onto any other DSP, RISC or general-purpose platform in a short time: 1-2 months.

Licensing Terms. To use the vocoder, customer should obtain a license from DSPINI only.

Customization. The vocoder can be customized under any specific requirements- other bit rate, frame size, any other robustness to channel errors, etc. Please contact with us for details.

Prospects. DSPINI is impoving and developing continuously a set of new vocoders with range from 300 bps up to 9600 bps, based on the TWELP technology.

Related Software. This vocoder may be effectively used in a bundle with other DSPINI's products:

  • Linear and acoustic echo cancellers,
  • Multichannel noise cancellers (including two-microphone adaptive array),
  • Wired or radiomodems for any types of channels and bitrates,
  • Other products.

Downloads: