EP0411360B1 - Method and apparatus for interference suppression in speech signals - Google Patents

Description

The invention is based on a method according to the type of the main claim and further comprises an arrangement for carrying out the method according to the invention.

Various methods are known for recognizing speech, but in which the recognition reliability is impaired by background noise. Such disturbing sound events occur, for example, in a moving motor vehicle in which the car radio is also switched on.

From the previously published report on the "36 TH IEE Vehicular Technology Conference", a method for reducing the interference of a telephone connection due to outside noise of the vehicle is known. According to this method, the external noise is recorded with a microphone attached outside the passenger compartment and its signals are linked with the microphone signal of the telephone handset in order to reduce noise. To reduce the interference of a telephone call by a car radio loudspeaker in the vehicle, it is also known from EP 56 587 to lower the sound level of the loudspeaker during the establishment of a telephone call connection.

The present invention relates to another field of recognition of voice signals, namely voice commands in a motor vehicle. The focus here is on the clear assignment of a command word from the driver or front passenger. Therefore, a reduction in the sound level of vehicle speakers is not satisfactory. Even when the level is lowered, words can appear in the program which, although undesirable, could act as command words.

This danger can be significantly reduced by the solution of the task according to the invention, which is characterized by the features of the main claim.

Arrangements required to carry out the method according to the invention can be implemented in a simple manner by means of modern semiconductor technology.

The measures listed in the subclaims enable advantageous developments and improvements of the invention specified in the main claim and advantageous arrangements for carrying out the method according to the invention.

Fig. 1

eine schematische Darstellung einer erfindungsgemäßen Anordnung in einem Kraftfahrzeug und

Fig. 2

ein Blockschaltbild einer erfindungsgemäßen Anordnung.

An embodiment of the invention is shown in the drawing using several figures and explained in more detail in the following description. It shows:

Fig. 1

a schematic representation of an arrangement according to the invention in a motor vehicle and

Fig. 2

a block diagram of an arrangement according to the invention.

Identical parts are provided with the same reference symbols in the figures.

In Fig. 1 only part of the passenger compartment 1 and the engine compartment 2 are shown of a motor vehicle. A microphone 3 is arranged in the passenger compartment, for example above the windshield, and picks up speech, driving noises and sound from a car radio 5, which is emitted by a loudspeaker 4. The sum signal Σ (t) emitted by the microphone 3 is fed to an arrangement 6 for carrying out the method according to the invention, at the output 7 of which the speech signal D₂ (t), which is largely free of interference, can be removed.

Another microphone 8 is arranged in the engine compartment 2, picks up driving noises and forwards them as a signal F (t) to the arrangement 6. Finally, the audio signal fed to the loudspeaker 4 is also fed to the arrangement 6 as signal A (t). If no monaural reproduction is provided on the car radio, a plurality of audio signals A (t) can be picked up and fed to the arrangement 6.

In the following considerations, the respective acoustic and electrical signals are equated. The signal picked up by the microphone 3 as a result of the sound radiation from the loudspeaker 4 differs from the signal A (t) by an initially unknown transfer function. The acting on the microphone 3 from the speaker 4 signal is therefore referred to as A₁ (t). In a similar manner, the driving noise F 1 (t) recorded by the microphone 3 differs from the driving noise F (t) recorded by the microphone 8. The sum signal thus results in Σ (t) = S (t) + F₁ (t) + A₁ (t).

The components of the sum signal F₁ (t) and A₁ (t) are A₁ (t) = H _A x A (t) and F₁ (t) = H _F x F (t),
where H _A and H _{F are} the corresponding transfer functions and x is a convolution.

In order to remove the components F₁ (t) and A₁ (t) from the sum signal, knowledge of the initially unknown functions H _F and H _{A is} required. Adaptive filters are used for this purpose, which are provided in the arrangement according to FIG. 2. The signals Σ (t) and F (t) are supplied by the microphones 3, 8 via suitable amplifiers 11, 12 and low-pass filters 13, 14 analog / digital converters 15, 16. The audio signal A (t) is also fed via an low-pass filter 17 to an analog / digital converter 18. The low-pass filters limit the bandwidth of the signals to a value that is necessary for the downstream speech recognition system.

A first adaptive filter 19 is used together with a correlator 20 to derive the signal A₁ (t) from the signal A (t). In a subtractor 21, the signal A₁ (t) is then subtracted from the sum signal Σ (t), whereby the signal D₁ (t) arises.

Another adaptive filter 22 and a correlator 23 are provided for deriving the signal F₁ (t) from the signal F (t). With the help of a further subtractor 24, the signal F 1 (t) is subtracted from the signal D 1 (t). At the output 7, the signal D₂ (t) is thus available for forwarding to a speech recognition system.

wobei ■ die Korrelationsfunktion bedeutet. Die Zielfunktion nutzt aus, daß für das richtige H_AZ das Signal [S(t) + F₁(t)] unkorreliert zum Audiosignal A(t) ist.The adaptive filter 19 is a non-recursive filter of approximately 100 order with the objective function $${\text{Z₁ (t) = (Σ (t) -H}}_{\text{AZ}}{\text{x A) ■ A (t) = 0 for H}}_{\text{AZ}}{\text{= H}}_{\text{A}}\text{,}$$

where ■ means the correlation function. The target function takes advantage of the fact that for the correct H _AZ the signal [S (t) + F₁ (t)] is uncorrelated to the audio signal A (t).

The filter coefficients can be calculated in detail using standard methods of digital signal processing, as described, for example, in the article "Adaptive Noise Canceling: Principles and Applications", Proceedings of the IEEE, Vol. 63, No. 12, December 1975, pages 1692 to 1716. The adaptive filters and the correlators can be implemented with suitably programmed signal processors.

The output signal D₁ (t) of the subtractor 21 is used both in the correlator 20 and supplied to the further subtractor 24. It results in D₁ (t) = Σ (t) - H _AZ x A (t).

To determine H _F , the signal D 1 (t), which is already freed from the audio signals, is used, which is composed of driving noise and speech. These are also uncorrelated, so that the method used for the audio signal can also be used for the noise signal. Here is the objective function $${\text{Z₂ (t) = [(D₁ (t) - H}}_{\text{FZ}}{\text{x F (t)] ■ F (t) = 0 for H}}_{\text{FZ}}{\text{= H}}_{\text{F}}\text{(t)}$$

The output signal of the subtractor 24 is then $${\text{D₂ (t) = S (t) + (H}}_{\text{F}}{\text{-H}}_{\text{FZ}}{\text{) xF (t) + [(H}}_{\text{A}}{\text{-H}}_{\text{AZ}}\text{) xA (t)] x F (t)}$$

From equation (3) it can be seen that the output signal D₂ (t) approaches the signal S (t), the better the approximation of H _FZ to H _F or H _AZ to H _A takes place. Even if the approach is incomplete, the interference component is significantly reduced.

Claims (5)

1. Speech-signal interference-elimination method for speech-recognition purposes in a motor vehicle, in which method the speech is converted, along with other simultaneously occurring sound events, into an electrical composite signal by means of a microphone (3) and a sound signal (f (t)) from a reference microphone (8) disposed outside the passenger compartment is fed via an adaptive filter (22) and then subtracted from the composite signal Σ (t), characterized in that control signals (A (t)) for a loudspeaker (4) disposed in the passenger compartment are fed in parallel with the composite signal Σ (t) via a further adaptive filter (19) and subtracted from the composite signal Σ (t), in that the resultant signal is compared with the control signal (A (t)) and in that the comparison signals obtained in this process control the further adaptive filter (19).
2. Method according to Claim 1, characterized in that the control signals are subjected to low-pass filtering and analogue/digital conversion upstream of the adaptive filter.
3. Arrangement for carrying out the method according to Claim 1, having a first microphone (3) inside the passenger compartment and a second microphone (8) outside the passenger compartment and a first adaptive filter (22) in the output of the second microphone, and having first means (24) for subtracting the output signal F₁ (t) of the adaptive filter (22) from the output signal D (t) of the first microphone (3), and having a correlator (23) for controlling the adaptive filter (22), characterized by a second adaptive filter (19) for control signals which control, in parallel therewith, a loudspeaker (4) disposed in the passenger compartment, by second means (21) for subtracting the output signal A₁ (t) of the second adaptive filter (19) from the output signal of the first microphone (3), which second means (21) are disposed upstream of the first means (24), and by a second correlator (20) which controls the second adaptive filter (19) and to which the control signals and the output signal of the second subtraction means (21) can be fed for the purpose of comparison.
4. Arrangement according to Claim 3, characterized in that low-pass filters (13, 14, 17) and A/D converters (15, 16, 18) are connected downstream of the microphones (3, 8) and the control-signal output for the loudspeaker (4).
5. Arrangement according to Claim 4, characterized in that the order of the adaptive filter is about 100.

Images