AMR-NB-WIP - Revision history

Superdump: /* IIR filtering */

2007-09-24T19:56:37Z

IIR filtering

← Older revision		Revision as of 12:56, 24 September 2007
Line 884:		Line 884:
	[[Image: Amrnb_pftilt.gif]]		[[Image: Amrnb_pftilt.gif]]

	~~; \mu~~ : ~~tilt factor equal to γ_t~~ . ~~k_1'~~		[[Image: Amrnb_mu.gif]]

	~~<insert summation form of filter>~~

	[[Image: Amrnb_reffac.gif]]		[[Image: Amrnb_reffac.gif]]
Line 894:		Line 892:
	; L_h : 22 - the truncation of the impulse response		; L_h : 22 - the truncation of the impulse response
	; h_f : impulse response of the formant filter H_f		; h_f : impulse response of the formant filter H_f

			So, in summation notation:

			[[Image: Amrnb_pftiltsum.gif]]

	===== 12.2 and 10.2 kbps modes =====		===== 12.2 and 10.2 kbps modes =====

Superdump: /* IIR filtering */

2007-09-24T19:41:48Z

IIR filtering

← Older revision		Revision as of 12:41, 24 September 2007
Line 878:		Line 878:
	^r(n) is filtered through 1/^A(z/γ_d).		^r(n) is filtered through 1/^A(z/γ_d).

	~~<insert summation form of filter>~~		[[Image: Amrnb_fmntfilt.gif]]

	The output is filtered through H_t(z) (the tilt compensation filter) resulting in the post-filtered speech signal, ^s_f(n).		The output is filtered through H_t(z) (the tilt compensation filter) resulting in the post-filtered speech signal, ^s_f(n).

Superdump: /* Calculate reconstructed speech samples */

2007-09-24T18:07:58Z

Calculate reconstructed speech samples

← Older revision		Revision as of 11:07, 24 September 2007
Line 849:		Line 849:
	; ^a_i : LP filter coefficients		; ^a_i : LP filter coefficients

	Note: for n = 0, ^s(0-i) should be take from previous speech samples if existing, else we will consider them 0 as this behaviour is undefined the specification.		Note: for n-i < 0, ^s(n-i) should be take from previous speech samples if existing, else we will consider them 0 as this behaviour is undefined the specification.

	== Additional instability protection ==		== Additional instability protection ==

Superdump: /* IIR filtering */

2007-09-21T19:37:12Z

IIR filtering

← Older revision		Revision as of 12:37, 21 September 2007
Line 893:		Line 893:

	; L_h : 22 - the truncation of the impulse response		; L_h : 22 - the truncation of the impulse response
			; h_f : impulse response of the formant filter H_f

	===== 12.2 and 10.2 kbps modes =====		===== 12.2 and 10.2 kbps modes =====

Superdump: /* IIR filtering */

2007-09-21T19:25:51Z

IIR filtering

← Older revision		Revision as of 12:25, 21 September 2007
Line 892:		Line 892:
	[[Image: Amrnb_r_h.gif]]		[[Image: Amrnb_r_h.gif]]

	L_h = 22		; L_h : 22 - the truncation of the impulse response

	===== 12.2 and 10.2 kbps modes =====		===== 12.2 and 10.2 kbps modes =====

Superdump: /* IIR filtering */

2007-09-21T19:24:31Z

IIR filtering

← Older revision		Revision as of 12:24, 21 September 2007
Line 891:		Line 891:

	[[Image: Amrnb_r_h.gif]]		[[Image: Amrnb_r_h.gif]]

			L_h = 22

	===== 12.2 and 10.2 kbps modes =====		===== 12.2 and 10.2 kbps modes =====

Superdump: /* IIR filtering */

2007-09-21T19:08:11Z

IIR filtering

← Older revision		Revision as of 12:08, 21 September 2007
Line 871:		Line 871:

	; γ_n, γ_d : control the amount of formant post-filtering		; γ_n, γ_d : control the amount of formant post-filtering

			The speech samples, ^s(n), are filtered through ^A(z/γ_n) to produce the residual signal ^r(n).

			[[Image: Amrnb_residual.gif]]

			^r(n) is filtered through 1/^A(z/γ_d).

			<insert summation form of filter>

			The output is filtered through H_t(z) (the tilt compensation filter) resulting in the post-filtered speech signal, ^s_f(n).

	[[Image: Amrnb_pftilt.gif]]		[[Image: Amrnb_pftilt.gif]]

	; \mu : tilt factor equal to γ_t . k_1'		; \mu : tilt factor equal to γ_t . k_1'

			<insert summation form of filter>

	[[Image: Amrnb_reffac.gif]]		[[Image: Amrnb_reffac.gif]]

	[[Image: Amrnb_r_h.gif]]		[[Image: Amrnb_r_h.gif]]

	The speech samples, ^s(n), are filtered through ^A(z/γ_n) to produce the residual signal ^r(n). ^r(n) is filtered through 1/^A(z/γ_d) and the output is filtered through H_t(z) (the tilt compensation filter) resulting in the post-filtered speech signal, ^s_f(n).

	===== 12.2 and 10.2 kbps modes =====		===== 12.2 and 10.2 kbps modes =====

Superdump: /* High-pass filtering and upscaling */

2007-09-21T17:25:05Z

High-pass filtering and upscaling

← Older revision		Revision as of 10:25, 21 September 2007
Line 913:		Line 913:

	(c.f. §6.2.2)		(c.f. §6.2.2)

			[[Image: Amrnb_highpass.gif]]

			After completing all filtering, the samples are scaled up by a factor of 2.

Superdump: /* Adaptive post-filtering */

2007-09-21T17:05:31Z

Adaptive post-filtering

← Older revision		Revision as of 10:05, 21 September 2007
Line 885:		Line 885:

	γ_n = 0.7		γ_n = 0.7

	γ_d = 0.75		γ_d = 0.75

	γ_t = 0.8 if k_1' > 0; 0 otherwise		γ_t = 0.8 if k_1' > 0; 0 otherwise

Line 891:		Line 893:

	γ_n = 0.55		γ_n = 0.55

	γ_d = 0.7		γ_d = 0.7

	γ_t = 0.8		γ_t = 0.8

Superdump: /* Adaptive post-filtering */

2007-09-21T16:54:01Z

Adaptive post-filtering

← Older revision		Revision as of 09:54, 21 September 2007
Line 863:		Line 863:

	(c.f. §6.2.1)		(c.f. §6.2.1)

			==== IIR filtering ====

	The speech samples, ^s(n), are filtered through a formant filter and a tilt compensation filter.		The speech samples, ^s(n), are filtered through a formant filter and a tilt compensation filter.
Line 880:		Line 882:
	The speech samples, ^s(n), are filtered through ^A(z/γ_n) to produce the residual signal ^r(n). ^r(n) is filtered through 1/^A(z/γ_d) and the output is filtered through H_t(z) (the tilt compensation filter) resulting in the post-filtered speech signal, ^s_f(n).		The speech samples, ^s(n), are filtered through ^A(z/γ_n) to produce the residual signal ^r(n). ^r(n) is filtered through 1/^A(z/γ_d) and the output is filtered through H_t(z) (the tilt compensation filter) resulting in the post-filtered speech signal, ^s_f(n).

	~~Q: How are {\gamma}_{[d\|n\|t]} found? Stored in a table?~~		===== 12.2 and 10.2 kbps modes =====

			γ_n = 0.7
			γ_d = 0.75
			γ_t = 0.8 if k_1' > 0; 0 otherwise

			===== Other modes =====

			γ_n = 0.55
			γ_d = 0.7
			γ_t = 0.8

			==== Adaptive gain control ====

	Adaptive gain control is used to compensate for the gain difference between the filtered and synthesised speech signals.		Adaptive gain control is used to compensate for the gain difference between the filtered and synthesised speech signals.
Line 891:		Line 905:

	; \alpha : adaptive gain control factor equal to 0.9		; \alpha : adaptive gain control factor equal to 0.9


	=== High-pass filtering and upscaling ===		=== High-pass filtering and upscaling ===

	(c.f. §6.2.2)		(c.f. §6.2.2)