WoodstitchHuman evals for voice AI

TTS voice leaderboard · US English · preliminary

Best TTS Voices for Phone Support

Stock text-to-speech voices ranked by blind paired listening on support-agent lines, heard over a phone connection.

Voices53 from 19 providersJudgments1,297 so farListeners71 paid, USUpdated21 Nov 2026, 21:00 UTCHow we evaluate →

Top 20 voices

Elo-scale strength from blind paired listening (1000 = average voice in this use case) · whiskers = 95% interval · higher is better

#RangeVoiceScoreEloJudgmentsPlay
11–32 KatieCartesia
1177±19050
21–30 DanielCartesia
1168±18447
31–39 RuthAmazon Polly
1135±19351
41–35 DaroGoogle Gemini TTS
1128±17251
51–35 HopeElevenLabs
1126±16952
61–37 MatthewAmazon Polly
1123±18157
71–43 MarkElevenLabs
1122±26049
81–37 KoreGoogle Gemini TTS
1113±17447
91–43 SavannahZonos3-Mini
1109±26023
101–39 KelseyInworld
1105±18154
112–40 MarinOpenAI
1087±15952
121–41 AshleyInworld
1087±19155
132–39 GeffenSpeechify
1085±18558
141–43 AutumnSmallest.ai
1083±18852
151–38 af_heartKokoro
1081±18444
162–38 ParkerCartesia
1079±15850
171–46 DefaultChatterbox
1075±25433
182–42 KelseySmallest.ai
1074±18554
193–40 HannahFish Audio
1067±16056
203–39 GordonMurf
1065±14850
212–39 FolaGoogle Gemini TTS
1064±15452
223–43 CharonGoogle Cloud TTS
1046±16149
236–42 AoedeGoogle Cloud TTS
1043±13855
244–41 LunaRime
1035±14951
252–45 GeorgeSpeechify
1035±18757
263–41 Jeff (M)Murf
1035±16954
274–44 Jessica Anne BogartElevenLabs
1029±18549
285–43 CelesteRime
1023±15250
291–48 af_bellaKokoro
1023±23535
307–45 RubySmallest.ai
1012±16752
3110–45 CedarOpenAI
1001±15263
327–44 AndromedaDeepgram
1001±14457
336–47 HenrySpeechify
997±17161
345–45 BiancaInworld
996±15851
355–48 AstraRime
989±19347
365–49 LeoOrpheus
972±23335
3714–48 JoannaAmazon Polly
958±16850
3813–52 TaraOrpheus
946±22635
3915–49 ArcasDeepgram
944±16655
4014–49 ThaliaDeepgram
944±15650
4115–49 VerityFish Audio
941±16758
429–51 Miles (M)Murf
940±20251
4314–51 OrusGoogle Cloud TTS
928±19451
4420–51 SarahFish Audio
894±19054
4527–52 Expressive narratorMiniMax
864±18551
4627–53 am_michaelKokoro
858±23233
4726–53 ZoeOrpheus
832±24442
4824–53 RyanQwen3-TTS
824±23938
4927–53 CoralOpenAI
822±21950
5038–53 DebatorMiniMax
760±22352
5143–53 PatientManMiniMax
749±19355
5243–53 AidenQwen3-TTS
711±20536
5339–53 WyattZonos3-Mini
696±25430

Listener question: “If you needed help with this issue, which voice would you prefer to hear from the support agent?” · 1,297 judgments from 71 listeners so far · 53 voices ranked; the rest are still collecting. Range = ranks consistent with the data at 95%. Phone clips pass through a narrowband phone condition; this is a listening test, not an agent benchmark. Methodology

Independent blind listening evaluation. Paid, vetted US listeners.
Methodology · Evaluated audio · Judgment data · Contact