SQLite Support in Phoneme Editor

synsinger

Jun 3

The phoneme editor has been rewritten to use SQLite:

Left clicking sets the left side of the phoneme, and right clicking sets the right. The arrow keys scroll through the list, and I can adjust the size of the segment with the left and right arrow keys.

That means that I can edit the phonemes with one hand on the mouse and the other on the arrow keys, which makes it fairly efficient.

Timing the recordings to a click track should simplify things, as the positions of phonemes would be locked in place and could inherit their timing instead of setting them all manually.

The reclist is now generated from a plaintext document that has simple macro replacements. For example, here's the portion where the CVCVC phonemes for /uh/ are declared. The !uh is a declaration of a macro value, and the @ is automatically replaced with the macro value. It's useful for common groups like this:

!uh @_plos 	p@-p@p 		-p_@, @_p, p_@, @_p- 	t@-t@t 		-t_@, @_t, t_@, @_t- 	k@-k@k 		-k_@, @_k, k_@, @_k- 	ch@-ch@ch	-ch_@, @_ch, ch_@, @_ch-  @_vplos 	b@-b@b 	-b_@, @_b, b_@, @_b- 	d@-d@d	-d_@, @_d, d_@, @_d- 	g@-g@g	-g_@, @_g, g_@, @_g- 	j@-j@j	-jh_@, @_jh, jh_@, @_jh-  @_nas 	m@-m@m		-m_@, @_m, m_@, @_m- 	n@-n@n		-n_@, @_n, n_@, @_n- 	ng@-ng@ng	-ng_@, @_ng, ng_@, @_ng- 	hh@-hh@hh	-hh_@, @_hh, hh_@, @_hh-  @_liq 	l@-l@l	-l_@, @_l, l_@, @_l- 	r@-r@r	-r_@, @_r, r_@, @_r- 	w@-w@w	-w_@, @_w, w_@, @_w- 	y@-y@y	-y_@, @_y, y_@, @_y-  @_fric 	f@-f@f	-f_@, @_f, f_@, @_f- 	s@-s@s	-s_@, @_s, s_@, @_s- 	sh@-sh@sh	-sh_@, @_sh, sh_@, @_sh- 	th@-th@th	-th_@, @_th, th_@, @_th-  @_vfric 	v@-v@v	-v_@, @_v, v_@, @_v- 	z@-z@z	-z_@, @_z, z_@, @_z- 	zh@-zh@zh	-zh_@, @_zh, zh_@, @_zh- 	dh@-dh@dh	-dh_@, @_dh, dh_@, @_dh-

And here's the declaration for /sp/ set of -CC phonemes:

sp_v-# 	speed	s_p_iy 	spin	s_p_ih 	sped	s_p_eh 	spak	s_p_ae # 	spot	s_p_aa 	spud	s_p_ah 	spall	s_p_ao 	spone	s_p_ow # 	spoor	s_p_uh 	spoon	s_p_uh 	spurt	s_p_er

I've only got about half of the reclist recorded, and I still need to add vowels.

But there's enough data in place that I can start working on the next step, which is analyzing the phonemes with WORLD, and storing their audio data into tables.

Once the phoneme audio data is in the database, I should be able to start synthesizing vocals.

At that point, I can listen to the results and start making additional decisions about how I'm editing the phonemes, and if I should start including additional data to enable smoother crossfading of phonemes.

Comment