Narrator bake-off

Temporary listening page · round 4B posted 7/29

Round 4B: a correction after the female mezzo and male scholar both sounded too sultry. The word “warm” and the emphasis on low pitch were removed. The four new Qwen briefs instead ask for medium pitch, crisp consonants, dry analytical authority, and public-lecture delivery. They explicitly prohibit breathiness, intimacy, sensuality, vocal fry, languor, and seductive pacing.

Tara stays here as the promising preset reference. Orpheus cannot add scholarly sharpness to that preset, so the new tests use Qwen VoiceDesign: two British female academics, one dry American woman professor, and one dry British male academic.

Round 4B · dry academic correction

Maineville micro-sample

Gemini 2.5 Flash · Iapetus (voice reference) 2.0 min · full note, not the 56-word local excerpt

Orpheus 3B 4-bit · Tara (promising reference) 21.3 sec · female preset; needs more scholarly sharpness

Qwen3-TTS 1.7B · British woman don 25.1 sec · clipped RP, dry and incisive

Qwen3-TTS 1.7B · British historian 22.2 sec · contemporary British, cool and lightly brisk

Qwen3-TTS 1.7B · American professor 23.8 sec · dry American control, crisp and unsentimental

Qwen3-TTS 1.7B · British male don 24.2 sec · medium pitch, clipped and impersonal

Round 4: five guarded local tests prompted by the Round 3 verdict. The local candidates all read the same 56-word Maineville passage. Qwen VoiceDesign auditions two deeper scholarly female designs and one male control, with explicit Charlus and Jupien pronunciation guidance. Orpheus auditions its Tara and Zoe female presets, which do not accept a style or pronunciation prompt.

The Gemini track is the existing full Maineville note, included as a voice reference rather than a same-text sample. MOSS was stopped when its real 17.6 GB model-plus-decoder footprint and throttled transfer breached the time guardrail. KugelAudio was gated out because the current MLX port requires 18.7 GB and ignores its voice selector.

Round 4 · deeper scholarly voices

Maineville micro-sample

Gemini 2.5 Flash · Iapetus (voice reference) 2.0 min · full note, not the 56-word local excerpt

Qwen3-TTS 1.7B VoiceDesign · female alto 25.0 sec · deep, mature humanities-professor brief

Qwen3-TTS 1.7B VoiceDesign · female mezzo 22.7 sec · rejected: too sultry

Orpheus 3B 4-bit · Tara 21.3 sec · promising, but needs more scholarly sharpness

Orpheus 3B 4-bit · Zoe 24.6 sec · female preset, no style prompt

Qwen3-TTS 1.7B VoiceDesign · male scholar 23.5 sec · rejected: too sultry

Round 3: four real “How to read this” notes, with the Gemini 2.5 Flash/Iapetus test voice first and free local models after it. Every engine reads the exact same title, summary, and explainer text. Kokoro uses a neutral American preset. Chatterbox Turbo uses its built-in stock voice, with no voice clone, reference recording, or style prompt. The older optional Qwen trial failed before producing audio and is not listed. Julia ruled Chatterbox “super off” and Kokoro “nothing like our Gemini voice.”

Round 3 · local annotation models

Spying on Morel at the Maineville brothel

Gemini 2.5 Flash · Iapetus (baseline) 2.0 min · American-register style prompt

Kokoro 82M · af_heart 2.0 min · neutral American preset, no style prompt

Chatterbox Turbo 4-bit · built-in stock voice 1.9 min · no style prompt, reference audio, or clone

Algebra at two in the morning

Gemini 2.5 Flash · Iapetus (baseline) 1.6 min · American-register style prompt

Kokoro 82M · af_heart 1.7 min · neutral American preset, no style prompt

Chatterbox Turbo 4-bit · built-in stock voice 1.5 min · no style prompt, reference audio, or clone

Why money has a colour

Gemini 2.5 Flash · Iapetus (baseline) 1.5 min · American-register style prompt

Kokoro 82M · af_heart 1.5 min · neutral American preset, no style prompt

Chatterbox Turbo 4-bit · built-in stock voice 1.4 min · no style prompt, reference audio, or clone

How jealousy enriches the mind

Gemini 2.5 Flash · Iapetus (baseline) 1.7 min · American-register style prompt

Kokoro 82M · af_heart 1.6 min · neutral American preset, no style prompt

Chatterbox Turbo 4-bit · built-in stock voice 1.5 min · no style prompt, reference audio, or clone

Round 2 changes: style prompt now demands a consistent American accent with French names in the “wine register” (Merlot: silent final consonant, English stress). Charon was too deep, so three lighter voices audition on P1. P3 appears both prompt-steered and lexicon-respelled (the model literally reads “Vair-duh-ran”, “Sant-oo-vairt”, “Frahnk-toh”) to compare the two levers.