Skip to content
TypeScript

feat(web): sort and filter transcription models (#32045)

Lakshmi Tanmayc09060b62bfe
diff --git a/apps/web/src/components/settings/VoiceSettingsPanel.tsx b/apps/web/src/components/settings/VoiceSettingsPanel.tsx
index c08844bafe0..b0fd5d2fff5 100644
--- a/apps/web/src/components/settings/VoiceSettingsPanel.tsx
+++ b/apps/web/src/components/settings/VoiceSettingsPanel.tsx
@@ -31,6 +31,7 @@ import {
MicIcon,
RefreshCwIcon,
StarIcon,
+ SlidersHorizontalIcon,
Trash2Icon,
XIcon,
} from "lucide-react";
@@ -65,6 +66,19 @@ import { Badge } from "../ui/badge";
import { Tooltip, TooltipPopup, TooltipTrigger } from "../ui/tooltip";
import { Button } from "../ui/button";
import { Input } from "../ui/input";
+import { InputGroup, InputGroupAddon, InputGroupInput } from "../ui/input-group";
+import {
+ Menu,
+ MenuCheckboxItem,
+ MenuGroup,
+ MenuGroupLabel,
+ MenuPopup,
+ MenuRadioGroup,
+ MenuRadioItem,
+ MenuRadioItemIndicator,
+ MenuSeparator,
+ MenuTrigger,
+} from "../ui/menu";
import { Select, SelectItem, SelectPopup, SelectTrigger, SelectValue } from "../ui/select";
import { Switch } from "../ui/switch";
import { Textarea } from "../ui/textarea";
@@ -72,6 +86,7 @@ import { toastManager } from "../ui/toast";
import { MicrophoneTest } from "./MicrophoneTest";
import { LanguageRow, ModelBenchmarks, orderSpeechLanguages } from "./SpeechModelComparison";
import { languageLabel, SpeechLanguageSelect } from "./SpeechLanguageSelect";
+import { filterSpeechModels, type SpeechModelSort } from "./speechModelBrowser";
import { AdvancedSettings } from "./AdvancedSettings";
import {
SETTINGS_PICKER_TRIGGER_CLASSNAME,
@@ -110,12 +125,6 @@ const spellingsOf = (words: SpeechCustomWords) =>
[term, ...aliases].map((spelling) => spelling.toLocaleLowerCase()),
),
);
-const modelSortOrder = (model: EnvironmentSpeechModel) => {
- if (model.active) return 0;
- if (model.state === "installed") return 1;
- if (model.state === "downloading" || model.state === "verifying") return 2;
- return model.recommended ? 3 : 4;
-};
type ModelCardProps = {
readonly model: EnvironmentSpeechModel;
@@ -717,6 +726,8 @@ function SpeechModelBrowser(props: {
catalog: { status, models, run, controlsDisabled, canCancel, requestedDownload },
} = props;
const [search, setSearch] = useState("");
+ const [sort, setSort] = useState<SpeechModelSort>("recommended");
+ const [downloadedOnly, setDownloadedOnly] = useState(false);
const [languageSearch, setLanguageSearch] = useState("");
const [selection, setSelection] = useState<string[] | null>(null);
const preferred = navigator.languages.map((language) => language.toLowerCase().split("-")[0]!);
@@ -729,14 +740,20 @@ function SpeechModelBrowser(props: {
(code) =>
!needle || code.includes(needle) || languageLabel(code).toLocaleLowerCase().includes(needle),
);
- const query = search.trim().toLocaleLowerCase();
- const sortedModels = models
- .filter(
- (model) =>
- selectedLanguages.every((language) => model.languages.includes(language)) &&
- (!query || (model.name + " " + model.description).toLocaleLowerCase().includes(query)),
- )
- .toSorted((left, right) => modelSortOrder(left) - modelSortOrder(right));
+ const accuracyLanguage = selectedLanguages[0] ?? "en";
+ const sortLabels: Record<SpeechModelSort, string> = {
+ recommended: "Recommended",
+ size: "Smallest download",
+ accuracy: `Best ${languageLabel(accuracyLanguage)} accuracy`,
+ speed: "Fastest transcription",
+ };
+ const sortedModels = filterSpeechModels(models, {
+ search,
+ languages: selectedLanguages,
+ downloadedOnly,
+ sort,
+ accuracyLanguage,
+ });
const activeModel = models.find((model) => model.active);
const renderedModels =
activeModel && !sortedModels.includes(activeModel)
@@ -808,13 +825,54 @@ function SpeechModelBrowser(props: {
? "Models for " + selectedLanguages.map(languageLabel).join(", ")
: "All models"}
</p>
- <Input
- size="compact"
- aria-label="Search transcription models"
- placeholder="Search models"
- value={search}
- onChange={(e) => setSearch(e.target.value)}
- />
+ <InputGroup>
+ <InputGroupInput
+ size="compact"
+ aria-label="Search transcription models"
+ placeholder="Search models"
+ value={search}
+ onChange={(e) => setSearch(e.target.value)}
+ />
+ <InputGroupAddon align="inline-end">
+ <Menu>
+ <MenuTrigger
+ render={<Button size="compact" variant="ghost-muted" />}
+ aria-label={`Sort and filter models: ${sortLabels[sort]}${downloadedOnly ? ", downloaded only" : ""}`}
+ >
+ <SlidersHorizontalIcon />
+ <span className="hidden @min-[40rem]/voice-models:inline">
+ {sortLabels[sort]}
+ </span>
+ {downloadedOnly && <span>1</span>}
+ </MenuTrigger>
+ <MenuPopup align="end">
+ <MenuGroup>
+ <MenuGroupLabel>Sort by</MenuGroupLabel>
+ <MenuRadioGroup
+ value={sort}
+ onValueChange={(value) => setSort(value as SpeechModelSort)}
+ >
+ {(Object.keys(sortLabels) as SpeechModelSort[]).map((value) => (
+ <MenuRadioItem key={value} value={value} closeOnClick>
+ <span className="flex items-center justify-between gap-3">
+ {sortLabels[value]}
+ <MenuRadioItemIndicator />
+ </span>
+ </MenuRadioItem>
+ ))}
+ </MenuRadioGroup>
+ </MenuGroup>
+ <MenuSeparator />
+ <MenuCheckboxItem
+ checked={downloadedOnly}
+ onCheckedChange={setDownloadedOnly}
+ >
+ Downloaded only
+ </MenuCheckboxItem>
+ </MenuPopup>
+ </Menu>
+ </InputGroupAddon>
+ </InputGroup>
</div>
<div
className="h-72 overflow-y-auto divide-y divide-border/60"
diff --git a/apps/web/src/components/settings/speechModelBrowser.test.ts b/apps/web/src/components/settings/speechModelBrowser.test.ts
new file mode 100644
index 00000000000..ad97b22a12a
--- /dev/null
+++ b/apps/web/src/components/settings/speechModelBrowser.test.ts
@@ -0,0 +1,100 @@
+import type { EnvironmentSpeechModel } from "@t3tools/contracts";
+import { expect, it } from "vite-plus/test";
+
+import { filterSpeechModels } from "./speechModelBrowser";
+
+const model: EnvironmentSpeechModel = {
+ id: "active",
+ name: "Active",
+ description: "Local transcription",
+ size: 300,
+ languages: ["en", "fr"],
+ accuracy: 1,
+ speed: 1,
+ recommended: false,
+ supportsStreaming: false,
+ supportsLanguageDetection: false,
+ supportsTranslation: false,
+ active: true,
+ state: "installed",
+};
+const models: readonly EnvironmentSpeechModel[] = [
+ model,
+ {
+ ...model,
+ id: "small",
+ name: "Small",
+ size: 100,
+ active: false,
+ state: "downloadable",
+ recommended: true,
+ benchmark: {
+ accuracy: {
+ en: { grade: "A", errorPercent: 3.8 },
+ fr: { grade: "B", errorPercent: 8 },
+ },
+ secondsPer30s: 2,
+ },
+ },
+ {
+ ...model,
+ id: "fast",
+ name: "Fast",
+ size: 200,
+ active: false,
+ benchmark: {
+ accuracy: {
+ en: { grade: "A", errorPercent: 3.2 },
+ fr: { grade: "B", errorPercent: 9 },
+ },
+ secondsPer30s: 1,
+ },
+ },
+ {
+ ...model,
+ id: "pending",
+ name: "Pending",
+ size: 400,
+ active: false,
+ state: "verifying",
+ languages: ["en"],
+ benchmark: { accuracy: {}, secondsPer30s: null },
+ },
+];
+
+it.each([
+ ["recommended", "en", ["active", "fast", "pending", "small"]],
+ ["size", "en", ["small", "fast", "active", "pending"]],
+ ["speed", "en", ["fast", "small", "active", "pending"]],
+ ["accuracy", "en", ["fast", "small", "active", "pending"]],
+ ["accuracy", "fr", ["small", "fast", "active", "pending"]],
+] as const)(
+ "sorts by %s for %s with missing benchmarks last",
+ (sort, accuracyLanguage, expected) => {
+ expect(
+ filterSpeechModels(models, {
+ search: "",
+ languages: [],
+ downloadedOnly: false,
+ sort,
+ accuracyLanguage,
+ }).map((entry) => entry.id),
+ ).toEqual(expected);
+ expect(models[0]?.id).toBe("active");
+ },
+);
+
+it("combines text, all selected languages, and completed downloads without retaining the active model", () => {
+ const options = {
+ search: " LOCAL ",
+ languages: ["en", "fr"],
+ downloadedOnly: true,
+ sort: "size",
+ accuracyLanguage: "en",
+ } as const;
+ expect(filterSpeechModels(models, options).map((entry) => entry.id)).toEqual(["fast", "active"]);
+ expect(
+ filterSpeechModels(models, { ...options, search: "fast" }).map((entry) => entry.id),
+ ).toEqual(["fast"]);
+ expect(filterSpeechModels(models, { ...options, search: "small" })).toEqual([]);
+});
diff --git a/apps/web/src/components/settings/speechModelBrowser.ts b/apps/web/src/components/settings/speechModelBrowser.ts
new file mode 100644
index 00000000000..d9cab8d1ef0
--- /dev/null
+++ b/apps/web/src/components/settings/speechModelBrowser.ts
@@ -0,0 +1,45 @@
+import type { EnvironmentSpeechModel } from "@t3tools/contracts";
+
+export type SpeechModelSort = "recommended" | "size" | "accuracy" | "speed";
+
+const modelSortOrder = (model: EnvironmentSpeechModel) => {
+ if (model.active) return 0;
+ if (model.state === "installed") return 1;
+ if (model.state === "downloading" || model.state === "verifying") return 2;
+ return model.recommended ? 3 : 4;
+};
+
+export function filterSpeechModels(
+ models: readonly EnvironmentSpeechModel[],
+ options: {
+ readonly search: string;
+ readonly languages: readonly string[];
+ readonly downloadedOnly: boolean;
+ readonly sort: SpeechModelSort;
+ readonly accuracyLanguage: string;
+ },
+) {
+ const query = options.search.trim().toLocaleLowerCase();
+ const score = (model: EnvironmentSpeechModel) => {
+ switch (options.sort) {
+ case "size":
+ return model.size;
+ case "accuracy":
+ return model.benchmark?.accuracy[options.accuracyLanguage]?.errorPercent ?? Infinity;
+ case "speed":
+ return model.benchmark?.secondsPer30s ?? Infinity;
+ case "recommended":
+ return modelSortOrder(model);
+ }
+ };
+ return models
+ .filter(
+ (model) =>
+ options.languages.every((language) => model.languages.includes(language)) &&
+ (!options.downloadedOnly || model.state === "installed") &&
+ (!query || (model.name + " " + model.description).toLocaleLowerCase().includes(query)),
+ )
+ .toSorted(
+ (left, right) => score(left) - score(right) || modelSortOrder(left) - modelSortOrder(right),
+ );
+}