VOICES.md
7.4 KB · 162 lines · markdown Raw
1 # Voices
2
3 - 🇺🇸 [American English](#american-english): 11F 9M
4 - 🇬🇧 [British English](#british-english): 4F 4M
5 - 🇯🇵 [Japanese](#japanese): 4F 1M
6 - 🇨🇳 [Mandarin Chinese](#mandarin-chinese): 4F 4M
7 - 🇪🇸 [Spanish](#spanish): 1F 2M
8 - 🇫🇷 [French](#french): 1F
9 - 🇮🇳 [Hindi](#hindi): 2F 2M
10 - 🇮🇹 [Italian](#italian): 1F 1M
11 - 🇧🇷 [Brazilian Portuguese](#brazilian-portuguese): 1F 2M
12
13 For each voice, the given grades are intended to be estimates of the **quality and quantity** of its associated training data, both of which impact overall inference quality.
14
15 Subjectively, voices will sound better or worse to different people.
16
17 Support for non-English languages may be absent or thin due to weak G2P and/or lack of training data. Some languages are only represented by a small handful or even just one voice (French).
18
19 Most voices perform best on a "goldilocks range" of 100-200 tokens out of ~500 possible. Voices may perform worse at the extremes:
20 - **Weakness** on short utterances, especially less than 10-20 tokens. Root cause could be lack of short-utterance training data and/or model architecture. One possible inference mitigation is to bundle shorter utterances together.
21 - **Rushing** on long utterances, especially over 400 tokens. You can chunk down to shorter utterances or adjust the `speed` parameter to mitigate this.
22
23 **Target Quality**
24 - How high quality is the reference voice? This grade may be impacted by audio quality, artifacts, compression, & sample rate.
25 - How well do the text labels match the audio? Text/audio misalignment (e.g. from hallucinations) will lower this grade.
26
27 **Training Duration**
28 - How much audio was seen during training? Smaller durations result in a lower overall grade.
29 - 10 hours <= **HH hours** < 100 hours
30 - 1 hour <= H hours < 10 hours
31 - 10 minutes <= MM minutes < 100 minutes
32 - 1 minute <= _M minutes_ 🤏 < 10 minutes
33
34 ### American English
35
36 - `lang_code='a'` in [`misaki[en]`](https://github.com/hexgrad/misaki)
37 - espeak-ng `en-us` fallback
38
39 | Name | Traits | Target Quality | Training Duration | Overall Grade | SHA256 |
40 | ---- | ------ | -------------- | ----------------- | ------------- | ------ |
41 | **af\_heart** | 🚺❤️ | | | **A** | `0ab5709b` |
42 | af_alloy | 🚺 | B | MM minutes | C | `6d877149` |
43 | af_aoede | 🚺 | B | H hours | C+ | `c03bd1a4` |
44 | af_bella | 🚺🔥 | **A** | **HH hours** | **A-** | `8cb64e02` |
45 | af_jessica | 🚺 | C | MM minutes | D | `cdfdccb8` |
46 | af_kore | 🚺 | B | H hours | C+ | `8bfbc512` |
47 | af_nicole | 🚺🎧 | B | **HH hours** | B- | `c5561808` |
48 | af_nova | 🚺 | B | MM minutes | C | `e0233676` |
49 | af_river | 🚺 | C | MM minutes | D | `e149459b` |
50 | af_sarah | 🚺 | B | H hours | C+ | `49bd364e` |
51 | af_sky | 🚺 | B | _M minutes_ 🤏 | C- | `c799548a` |
52 | am_adam | 🚹 | D | H hours | F+ | `ced7e284` |
53 | am_echo | 🚹 | C | MM minutes | D | `8bcfdc85` |
54 | am_eric | 🚹 | C | MM minutes | D | `ada66f0e` |
55 | am_fenrir | 🚹 | B | H hours | C+ | `98e507ec` |
56 | am_liam | 🚹 | C | MM minutes | D | `c8255075` |
57 | am_michael | 🚹 | B | H hours | C+ | `9a443b79` |
58 | am_onyx | 🚹 | C | MM minutes | D | `e8452be1` |
59 | am_puck | 🚹 | B | H hours | C+ | `dd1d8973` |
60 | am_santa | 🚹 | C | _M minutes_ 🤏 | D- | `7f2f7582` |
61
62 ### British English
63
64 - `lang_code='b'` in [`misaki[en]`](https://github.com/hexgrad/misaki)
65 - espeak-ng `en-gb` fallback
66
67 | Name | Traits | Target Quality | Training Duration | Overall Grade | SHA256 |
68 | ---- | ------ | -------------- | ----------------- | ------------- | ------ |
69 | bf_alice | 🚺 | C | MM minutes | D | `d292651b` |
70 | bf_emma | 🚺 | B | **HH hours** | B- | `d0a423de` |
71 | bf_isabella | 🚺 | B | MM minutes | C | `cdd4c370` |
72 | bf_lily | 🚺 | C | MM minutes | D | `6e09c2e4` |
73 | bm_daniel | 🚹 | C | MM minutes | D | `fc3fce4e` |
74 | bm_fable | 🚹 | B | MM minutes | C | `d44935f3` |
75 | bm_george | 🚹 | B | MM minutes | C | `f1bc8122` |
76 | bm_lewis | 🚹 | C | H hours | D+ | `b5204750` |
77
78 ### Japanese
79
80 - `lang_code='j'` in [`misaki[ja]`](https://github.com/hexgrad/misaki)
81 - Total Japanese training data: H hours
82
83 | Name | Traits | Target Quality | Training Duration | Overall Grade | SHA256 | CC BY |
84 | ---- | ------ | -------------- | ----------------- | ------------- | ------ | ----- |
85 | jf_alpha | 🚺 | B | H hours | C+ | `1bf4c9dc` | |
86 | jf_gongitsune | 🚺 | B | MM minutes | C | `1b171917` | [gongitsune](https://github.com/koniwa/koniwa/blob/master/source/tnc/tnc__gongitsune.txt) |
87 | jf_nezumi | 🚺 | B | _M minutes_ 🤏 | C- | `d83f007a` | [nezuminoyomeiri](https://github.com/koniwa/koniwa/blob/master/source/tnc/tnc__nezuminoyomeiri.txt) |
88 | jf_tebukuro | 🚺 | B | MM minutes | C | `0d691790` | [tebukurowokaini](https://github.com/koniwa/koniwa/blob/master/source/tnc/tnc__tebukurowokaini.txt) |
89 | jm_kumo | 🚹 | B | _M minutes_ 🤏 | C- | `98340afd` | [kumonoito](https://github.com/koniwa/koniwa/blob/master/source/tnc/tnc__kumonoito.txt) |
90
91 ### Mandarin Chinese
92
93 - `lang_code='z'` in [`misaki[zh]`](https://github.com/hexgrad/misaki)
94 - Total Mandarin Chinese training data: H hours
95
96 | Name | Traits | Target Quality | Training Duration | Overall Grade | SHA256 |
97 | ---- | ------ | -------------- | ----------------- | ------------- | ------ |
98 | zf_xiaobei | 🚺 | C | MM minutes | D | `9b76be63` |
99 | zf_xiaoni | 🚺 | C | MM minutes | D | `95b49f16` |
100 | zf_xiaoxiao | 🚺 | C | MM minutes | D | `cfaf6f2d` |
101 | zf_xiaoyi | 🚺 | C | MM minutes | D | `b5235dba` |
102 | zm_yunjian | 🚹 | C | MM minutes | D | `76cbf8ba` |
103 | zm_yunxi | 🚹 | C | MM minutes | D | `dbe6e1ce` |
104 | zm_yunxia | 🚹 | C | MM minutes | D | `bb2b03b0` |
105 | zm_yunyang | 🚹 | C | MM minutes | D | `5238ac22` |
106
107 ### Spanish
108
109 - `lang_code='e'` in [`misaki[en]`](https://github.com/hexgrad/misaki)
110 - espeak-ng `es`
111
112 | Name | Traits | SHA256 |
113 | ---- | ------ | ------ |
114 | ef_dora | 🚺 | `d9d69b0f` |
115 | em_alex | 🚹 | `5eac53f7` |
116 | em_santa | 🚹 | `aa8620cb` |
117
118 ### French
119
120 - `lang_code='f'` in [`misaki[en]`](https://github.com/hexgrad/misaki)
121 - espeak-ng `fr-fr`
122 - Total French training data: <11 hours
123
124 | Name | Traits | Target Quality | Training Duration | Overall Grade | SHA256 | CC BY |
125 | ---- | ------ | -------------- | ----------------- | ------------- | ------ | ----- |
126 | ff_siwis | 🚺 | B | <11 hours | B- | `8073bf2d` | [SIWIS](https://datashare.ed.ac.uk/handle/10283/2353) |
127
128 ### Hindi
129
130 - `lang_code='h'` in [`misaki[en]`](https://github.com/hexgrad/misaki)
131 - espeak-ng `hi`
132 - Total Hindi training data: H hours
133
134 | Name | Traits | Target Quality | Training Duration | Overall Grade | SHA256 |
135 | ---- | ------ | -------------- | ----------------- | ------------- | ------ |
136 | hf_alpha | 🚺 | B | MM minutes | C | `06906fe0` |
137 | hf_beta | 🚺 | B | MM minutes | C | `63c0a1a6` |
138 | hm_omega | 🚹 | B | MM minutes | C | `b55f02a8` |
139 | hm_psi | 🚹 | B | MM minutes | C | `2f0f055c` |
140
141 ### Italian
142
143 - `lang_code='i'` in [`misaki[en]`](https://github.com/hexgrad/misaki)
144 - espeak-ng `it`
145 - Total Italian training data: H hours
146
147 | Name | Traits | Target Quality | Training Duration | Overall Grade | SHA256 |
148 | ---- | ------ | -------------- | ----------------- | ------------- | ------ |
149 | if_sara | 🚺 | B | MM minutes | C | `6c0b253b` |
150 | im_nicola | 🚹 | B | MM minutes | C | `234ed066` |
151
152 ### Brazilian Portuguese
153
154 - `lang_code='p'` in [`misaki[en]`](https://github.com/hexgrad/misaki)
155 - espeak-ng `pt-br`
156
157 | Name | Traits | SHA256 |
158 | ---- | ------ | ------ |
159 | pf_dora | 🚺 | `07e4ff98` |
160 | pm_alex | 🚹 | `cf0ba8c5` |
161 | pm_santa | 🚹 | `d4210316` |
162