fix(tts): forward the OpenAI speed field to the backend (#11097) (#11120)

* fix(tts): forward the OpenAI speed field to the backend (#11097)

/v1/audio/speech accepted the documented OpenAI `speed` field and then
dropped it: schema.TTSRequest had no Speed member, so the value never
reached proto.TTSRequest and the request returned 200 with an unchanged
playback rate.

Accept speed and normalise it into the existing per-request params map,
which core/backend forwards verbatim to the backend. An explicit
params["speed"] still wins, and a value outside the documented 0.25-4.0
range is now rejected with 400 instead of being silently ignored.

Signed-off-by: Anai-Guo <antai12232931@outlook.com>

* fix(tts): distinguish explicit speed=0 from an omitted field

Make TTSRequest.Speed a *float32 so an explicit `"speed": 0` (invalid,
below the documented 0.25 minimum) is rejected with 400 instead of being
treated as unset and silently defaulted. An omitted field stays nil and
leaves the backend default untouched.

Add a request-boundary regression that distinguishes an omitted speed from
an explicit zero, addressing review feedback.

Signed-off-by: Anai-Guo <antai12232931@outlook.com>

* docs: drop the speed field from the TTS docs

Per review: no backend consumes params.speed today, so documenting it
would be misleading. The API-level plumbing and validation stay.

Signed-off-by: Anai-Guo <antai12232931@outlook.com>

---------

Signed-off-by: Anai-Guo <antai12232931@outlook.com>

Author: Anai-Guo

core/http/endpoints/localai/tts.go

@@ -40,6 +40,10 @@ func TTSEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, appConfig
 			return echo.ErrBadRequest
 		}
 
+		if err := applyTTSSpeed(input); err != nil {
+			return err
+		}
+
 		xlog.Debug("LocalAI TTS Request received", "model", input.Model)
 
 		if cfg.Backend == "" && input.Backend != "" {

core/http/endpoints/localai/tts_speed.go

@@ -0,0 +1,46 @@
+package localai
+
+import (
+	"fmt"
+	"net/http"
+	"strconv"
+
+	"github.com/labstack/echo/v4"
+	"github.com/mudler/LocalAI/core/schema"
+)
+
+// ttsSpeedMin and ttsSpeedMax mirror the range the OpenAI Speech API documents
+// for the `speed` field.
+const (
+	ttsSpeedMin = 0.25
+	ttsSpeedMax = 4.0
+)
+
+// applyTTSSpeed normalises the OpenAI `speed` field onto the backend-specific
+// params map, which is what actually reaches the gRPC TTSRequest. Without this
+// the field is parsed off the request and then dropped, so a call with
+// speed=0.8 returns HTTP 200 with an unchanged playback rate instead of either
+// honouring or rejecting it (#11097).
+//
+// An explicit `"speed": 0` is invalid (below the documented minimum) and is
+// rejected with 400 rather than treated as unset, which is why Speed is a
+// pointer. An explicit params["speed"] wins so the LocalAI-native extension keeps
+// precedence over the OpenAI-compatible field. Backends whose model exposes no
+// rate control ignore the param, exactly like they ignore `instructions`.
+func applyTTSSpeed(input *schema.TTSRequest) error {
+	if input.Speed == nil {
+		return nil
+	}
+	speed := *input.Speed
+	if speed < ttsSpeedMin || speed > ttsSpeedMax {
+		return echo.NewHTTPError(http.StatusBadRequest,
+			fmt.Sprintf("speed must be between %g and %g", ttsSpeedMin, ttsSpeedMax))
+	}
+	if input.Params == nil {
+		input.Params = make(map[string]string)
+	}
+	if _, ok := input.Params["speed"]; !ok {
+		input.Params["speed"] = strconv.FormatFloat(float64(speed), 'g', -1, 32)
+	}
+	return nil
+}

core/http/endpoints/localai/tts_speed_test.go

@@ -0,0 +1,86 @@
+package localai
+
+import (
+	"net/http"
+
+	"github.com/labstack/echo/v4"
+	"github.com/mudler/LocalAI/core/schema"
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+// f32 returns a pointer to v, matching how the OpenAI `speed` field is decoded
+// (a pointer so an explicit zero is distinguishable from an omitted field).
+func f32(v float32) *float32 { return &v }
+
+// Regression for #11097: /v1/audio/speech accepted the documented OpenAI
+// `speed` field, dropped it before the request reached the backend, and
+// returned 200 with an unchanged playback rate. The field must now be
+// normalised onto the params map that is forwarded to the gRPC TTSRequest,
+// and an out-of-range value must be rejected instead of silently ignored.
+var _ = Describe("applyTTSSpeed", func() {
+	It("forwards speed onto the backend params map", func() {
+		input := &schema.TTSRequest{Speed: f32(0.8)}
+		Expect(applyTTSSpeed(input)).To(Succeed())
+		Expect(input.Params).To(HaveKeyWithValue("speed", "0.8"))
+	})
+
+	It("leaves params untouched when speed is unset", func() {
+		input := &schema.TTSRequest{}
+		Expect(applyTTSSpeed(input)).To(Succeed())
+		Expect(input.Params).To(BeNil())
+	})
+
+	It("treats an omitted speed as unset but rejects an explicit zero", func() {
+		omitted := &schema.TTSRequest{}
+		Expect(applyTTSSpeed(omitted)).To(Succeed())
+		Expect(omitted.Params).To(BeNil())
+
+		explicitZero := &schema.TTSRequest{Speed: f32(0)}
+		err := applyTTSSpeed(explicitZero)
+		Expect(err).To(HaveOccurred())
+		httpErr, ok := err.(*echo.HTTPError)
+		Expect(ok).To(BeTrue())
+		Expect(httpErr.Code).To(Equal(http.StatusBadRequest))
+		Expect(explicitZero.Params).To(BeNil())
+	})
+
+	It("keeps an explicit params entry over the OpenAI field", func() {
+		input := &schema.TTSRequest{Speed: f32(0.8), Params: map[string]string{"speed": "1.5"}}
+		Expect(applyTTSSpeed(input)).To(Succeed())
+		Expect(input.Params).To(HaveKeyWithValue("speed", "1.5"))
+	})
+
+	It("preserves unrelated params", func() {
+		input := &schema.TTSRequest{Speed: f32(2), Params: map[string]string{"ref_text": "hello"}}
+		Expect(applyTTSSpeed(input)).To(Succeed())
+		Expect(input.Params).To(HaveKeyWithValue("ref_text", "hello"))
+		Expect(input.Params).To(HaveKeyWithValue("speed", "2"))
+	})
+
+	DescribeTable("rejects values outside the OpenAI range",
+		func(speed float32) {
+			input := &schema.TTSRequest{Speed: f32(speed)}
+			err := applyTTSSpeed(input)
+			Expect(err).To(HaveOccurred())
+			httpErr, ok := err.(*echo.HTTPError)
+			Expect(ok).To(BeTrue())
+			Expect(httpErr.Code).To(Equal(http.StatusBadRequest))
+			Expect(input.Params).To(BeNil())
+		},
+		Entry("below the minimum", float32(0.1)),
+		Entry("above the maximum", float32(4.5)),
+		Entry("negative", float32(-1)),
+		Entry("explicit zero (distinct from an omitted field)", float32(0)),
+	)
+
+	DescribeTable("accepts the documented bounds",
+		func(speed float32, want string) {
+			input := &schema.TTSRequest{Speed: f32(speed)}
+			Expect(applyTTSSpeed(input)).To(Succeed())
+			Expect(input.Params).To(HaveKeyWithValue("speed", want))
+		},
+		Entry("minimum", float32(0.25), "0.25"),
+		Entry("maximum", float32(4), "4"),
+	)
+})

core/schema/localai.go

@@ -80,6 +80,12 @@ type TTSRequest struct {
 	Format   string `json:"response_format,omitempty" yaml:"response_format,omitempty"` // (optional) output format
 	Stream     bool   `json:"stream,omitempty" yaml:"stream,omitempty"`                   // (optional) enable streaming TTS
 	SampleRate int    `json:"sample_rate,omitempty" yaml:"sample_rate,omitempty"`         // (optional) desired output sample rate
+	// Speed is the OpenAI `speed` field (0.25-4.0). It is a pointer so an
+	// explicit `"speed": 0` (invalid, rejected with 400) is distinguishable
+	// from an omitted field (left at the backend default). It is normalised
+	// into Params["speed"] so it reaches the backend over the same channel as
+	// the other per-request generation parameters.
+	Speed *float32 `json:"speed,omitempty" yaml:"speed,omitempty"`
 	// Instructions is a free-form, per-request style/voice description. It maps to
 	// the OpenAI `instructions` field and is forwarded to the backend so expressive
 	// TTS models (e.g. Qwen3-TTS CustomVoice/VoiceDesign) can vary tone or designed